# An Accurate Bilinear Cavern Model for Compressed Air Energy Storage

An Accurate Bilinear Cavern Model for Compressed Air Energy Storage https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 1 An Accurate Bilinear Cavern Model for Compressed Air Energy Storage a,b a a a Junpeng Zhan , Osama Aslam Ansari , Weijia Liu , C. Y. Chung Department of Electrical and Computer Engineering, University of Saskatchewan, Saskatoon, SK S7N 5A9, Canada Sustainable Energy Technologies Department, Brookhaven National Laboratory, Upton, NY 11973, U.S.A. Mass of air in virtual container 2 as shown in Fig. 3 Abstract—Compressed air energy storage is suitable for large- (kg) scale electrical energy storage, which is important for integrating renewable energy sources into electric power systems. A typical Mass of air charged into the cavern (kg) compressed air energy storage plant consists of compressors, Number of time steps expanders, caverns, and a motor/generator set. Current cavern Pressure of the air charged into the cavern (bar) models used for compressed air energy storage are either accurate Pressures in virtual states as shown in Fig. 2, = 1,2 but highly nonlinear or linear but inaccurate. The application of (bar) highly nonlinear cavern models in power system optimization Maximum and minimum pressures in a cavern for problems renders them computationally challenging to solve. In optimal operation of compressed air energy storage this regard, an accurate bilinear cavern model for compressed air (CAES), respectively (bar) energy storage is proposed in this paper. The charging and discharging processes in a cavern are divided into several Pressure of the air in the cavern after the charging, real/virtual states. The first law of thermodynamics and ideal gas discharging, and idle processes for =2, 3, and 4, law are then utilized to derive a cavern model, i.e., a model for the respectively (bar) variation of temperature and pressure in these processes. Time step Thereafter, the heat transfer between the air in the cavern and the Surface area of the cavern wall (m ) cavern wall is considered and integrated into the cavern model. By , Operational costs of charging and discharging power subsequently eliminating several negligible terms, the cavern in CAES, respectively (\$/MWh) model reduces to a bilinear model. The accuracy of the bilinear Maximum and minimum charging power of CAES, cavern model is verified via comparison with both an accurate respectively (MW) nonlinear model and two sets of field-measured data. The bilinear cavern model can be easily linearized and is then suitable for Maximum and minimum discharging power of integration into optimization problems considering compressed CAES, respectively (MW) air energy storage. This is verified via comparatively solving a self- Total internal energy (J) scheduling problem of compressed air energy storage using Specific air constant (J/(kg K)) different cavern models. Temperature of the air in the cavern after the Index Terms—Bilinear cavern model; compressed air energy charging, discharging, and idle processes for =2, 3, storage (CAES); heat transfer; ideal gas law; thermodynamics. and 4, respectively (K) Temperature of the cavern wall (K) Temperature of the air charged into the cavern (K) NOMENCLATURE Temperatures in virtual states as shown in Fig. 2, = Parameters 1,2 (K) a,ht Both adiabatic process and heat transfer are , Volumes of virtual containers as shown in Figs. 2 and considered (superscript) 3 (m ) Parameters, = 1,2,3, ⋯ ,13, given in the Appendix Volume of a cavern (m ) / Coefficients used to model the relationship between Work (J) charging/discharging power and mass flow rate in/out Average air density in a cavern (kg/m ) Parameters, = 0,1, representing the left-hand side Electricity price (\$/MWh) of (3) and (10), respectively, given in the Appendix Δ Time interval (s) Parameters, = 2,3, ⋯ ,32, given in the Appendix Δ Change in internal energy (J) Constant volume specific heat (J/(kg K)) 0,1,2, ⋯ , ( − 1) ht Heat transfer (superscript) Ω 1,2,3, ⋯ , Heat transfer coefficient (W/(m K)) A constant equal to 1.4 Variables Parameters used in (35), (36), (47), and (48), = 1,2,3,4 ̇ Mass flow rate charged into a cavern (kg/s) ̇ Mass flow rate discharged out of a cavern (kg/s) Mass of air in the cavern (kg) E-mail addresses: zhanjunpeng@gmail.com, oa.ansari@usask.ca, liuweijiamarcel@gmail.com, c.y.chung@usask.ca). https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 2 Pressure of air in the cavern (bar) about 75% of the continental U.S. has geologic sites suitable for ( ) CAES [7], [8]. Northern Europe is also replete with suitable salt Pressure after a (xx) process, where (xx) can be ‘ch’, ,() ‘dch’, and ‘idl’, which represent charging, deposits. For example, nearly 500 salt caverns are currently being used for natural gas storage. That is, it is feasible to install discharging, and idle, respectively (bar) Temperature of air in the cavern (K) CAES in many different locations, which makes it a promising ( ) large-scale energy storage technology. Temperature after a (xx) process, where (xx) can be ,() ‘ch’, ‘dch’, and ‘idl’, which represent charging, Discharg ing circuit discharging, and idle, respectively (K) Charging circuit , Charging and discharging power, respectively (MW) Binary variable indicating the charging and M/G 3 4 6 5 discharging processes, respectively 2 1 2 I. INTRODUCTION Air Fuel 8 Fuel Air Energy storage technologies are critical to electric power Air 1: Motor or Generator 2: Clutches systems, especially considering that the penetration of 3: High pressure turbine Charging Discharging renewable generation is growing rapidly, e.g., the share of wind 4: Low pressure turbine process process 5: Low pressure compressor power in global electricity generation will increase from 4% in 6: High pressure compresso r 2015 to 25-28% in 2050 [1]. Energy storage can provide various 7: Cavern (also represents Repository containers 3 or 5 in Fig. 2 kinds of services [2], [3], e.g., electric energy time-shift, electric (underground cavern) and container 3 in Fig. 3) supply capacity, regulation, power reliability, etc. The current 8: Represent virtual container 1 in Fig. 2 global installed energy storage capacity is approximately 141 GW, and an estimated 310 GW of additional capacity is needed Fig. 1. Schematic of the Huntorf compressed air energy storage plant. in the United States, Europe, China, and India [4] to support the massive increase of renewable generation in the future. Two Two commercialized CAES plants are currently in operation. The world’s first CAES plant was installed in Huntorf, types of large-scale energy storage currently exist, i.e., pumped- hydro power storage and compressed air energy storage Germany in 1978 [6]. The power ratings of its charging and discharging processes are 60 and 290 MW, respectively. The (CAES), which can be cost-effectively installed at the electricity grid scale. Hydropower technologies are mature, and second commercialized CAES plant is the McIntosh plant in McIntosh, Alabama, U.S. [8], which became operational in many sites that were feasible for constructing pumped-storage hydropower already have hydropower plants in place [5]. 1991. The power rating of its charging process is 50 MW and it can produce an output of 110 MW electricity for up to 26 hours. Therefore, this paper focuses on CAES technology. A CAES plant consists of compressors, turbines, a Due to several benefits offered by CAES, a number of CAES plants are being constructed for efficient utilization of motor/generator set, and large repositories, e.g., underground salt caverns. Fig. 1 shows a schematic of the Huntorf CAES renewable energy sources. For example, a feasibility study for a 160 MW CAES plant near the Saskatchewan-Alberta border plant to depict the main components of CAES [6]. CAES operates in one of three modes, i.e., charging, discharging, and in Canada [9] was finished in late 2018. This proposed CAES plant is expected to be combined with the interconnection idle. When charging, the motor/generator set acts as a motor that is connected to the compressors via a clutch, as shown in between the Saskatchewan and Alberta power grids. From 2009 to 2013, Pacific Gas & Electric received US\$50 million in the ‘Charging circuit’ part of Fig. 1. Low-cost electricity (e.g., off-peak electricity or wind power) is usually used by the motor funding for a demonstration project to validate the design, performance, and reliability of a 300 MW CAES plant in Kern to compress air to high pressure for storage in a large repository. When discharging, the motor/generator set acts as a generator County, California [10]. Several further examples are provided in [10], [11]. that is connected to the turbines via another clutch, as shown in the ‘Discharging circuit’ part of Fig. 1. The compressed air is The two existing CAES plants in Huntorf and Alabama are diabatic CAES (D-CAES), and also referred to as the first released from the repository and then combusted with fuel (e.g., natural gas) to drive the turbines. When idling, no air is injected generation CAES. Second generation CAES include adiabatic CAES (A-CAES) [12] [13], [14], [15] [16], [17], [18], into or released from the repository. Multiple stages of compressors (turbines) are typically used instead of a single isothermal CAES (I-CAES) [19], [20], and hybrid CAES [21], [22]. In the D-CAES, the heat generated in the compression stage to increase the efficiency, e.g., low- and high-pressure compressors (turbines) are used in the Huntorf CAES plant as process is wasted and an external heat source is required in the expansion process. In A-CAES, the heat generated in the shown in Fig. 1. CAES is a high power and high energy storage technology compression process is captured, stored, and then used in the expansion process. I-CAES prevents the temperature from and has relatively low capital, operational, and maintenance costs [3]. The power rating of a large-scale CAES plant can changing in the compression and expansion devices. Pacific Northwest National Laboratory (PNNL) and EPRI consider A- reach 300 or even 1000 MW and the rated energy capacity can reach 1000 or even 2860 MWh [3]. Different definitions of CAES and hybrid CAES to be the most suitable and promising CAES technologies [10]. For a complete review of CAES efficiencies for CAES are discussed in [7], e.g., the round-trip efficiency of the CAES typically ranges from 66 to 82%. technologies, readers are referred to [10], which provides a comprehensive classification and comparison of different According to the Electric Power Research Institute (EPRI), https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 3 CAES technologies as well as several challenging issues value terms and using several linearization methods such as relevant for CAES research and development. Newton’s generalized binomial theorem and the first-order Air reservoir is a fundamental part of the CAES as it Taylor series approximation [36]. This original bilinear model significantly affects the power and energy rating as well as the is derived step-by-step in this paper. The advantages of the total construction and operation costs [11]. The reservoirs can bilinear model over the existing model types mentioned above be above-ground containers or underground caverns, including are twofold: 1) it is accurate, as will be verified in this paper, salt, hard rock, and porous rock layer caverns [11], [23]. The and 2) it is suitable for integration in power system optimization existing commercial CAES plants mentioned above both use problems as explained in the following two paragraphs. underground salt caverns for air storage. The proposed cavern To consider CAES in an optimization problem, binary model is general and can be applied to the different kinds of variables are required to indicate the charging, discharging, or reservoirs mentioned above. idle status. That is, an optimization problem considering CAES Some research has been conducted to model compressors and is a mixed-integer programming problem. If the nonlinear turbines [24], [25], [26]. However, the CAES cavern models cavern models given in [27], [28], [29] are used for CAES, an available in the literature [11], [27], [28], [29], [30] are either optimization problem considering CAES becomes a mixed- too complicated or inaccurate and, as such, are not suitable for integer nonlinear programming (MINLP) problem. MINLP real applications. Therefore, this paper focuses on the cavern problems are usually difficult to be accurately solved in an modeling of CAES, i.e., modeling the temperature and pressure acceptable computational time [37], [38], and therefore variations of cavern air. Thermodynamic properties, such as nonlinear cavern models such as those given in [27], [28], [29] variations of temperature and pressure in the caverns and the are not suitable to be employed in an optimization problem. If heat transfer between caverns and their surroundings, are the proposed bilinear cavern model is used for CAES, an important factors that affect the overall plant operation and optimization problem considering CAES becomes a mixed- performance [27], [28], [29]. Two kinds of cavern models for integer bilinear programming (MIBLP) problem, which can CAES are currently described in the literature. easily be converted into a mixed-integer linear programming The first kind is models that are accurate but highly nonlinear (MILP) problem [39], [40]. MILP problems are relatively easier [27], [28], [29]. In [27], complex and simplified real gas models to solve than MINLP problems [37], [41]. MILP problems can are developed for an adiabatic cavern for CAES, both of which also be efficiently solved by several mature, off-the-shelf adequately represent the thermodynamic properties of the air. commercial solvers, e.g., CPLEX [42] and Gurobi [43]. Reference [28] developed an accurate dynamic simulation Therefore, the bilinear cavern model is more suitable for model for a CAES cavern that incorporates an accurate heat integration into an optimization problem compared to the transfer model; the heat transfer is shown to play an important highly nonlinear cavern models given in [27], [28], [29]. The role in the thermodynamic behavior of the cavern and therefore above discussion serves as the main motivation to propose the the model in [28] can accurately simulate the actual cavern bilinear model. behavior. In [29], a simplified and unified analytical solution A typical example of the previously mentioned optimization considering heat transfer is proposed for temperature and problem is as follows. A CAES plant can act as an independent pressure variations in CAES caverns. This model is validated participator in electricity markets. In this case, the self- using data from the Huntorf plant trial tests and the results scheduling of a CAES plant is an important optimization calculated from the models in [27] and [28], demonstrating that problem that needs to be solved to maximize its arbitrage the solution in [29] is capable of adequately calculating the revenue obtained from selling electricity to and buying thermodynamic behavior of CAES caverns. Note that the electricity from the market [34], [44], [45]. In this paper, the models described in [27], [28], and [29] are accurate but highly day-ahead self-scheduling problem (SSP) of CAES is used to nonlinear. demonstrate the advantage of using the proposed bilinear model The second kind of cavern model assumes that the air in an optimization problem compared to the highly nonlinear temperature in the cavern is constant. This kind of model has cavern models given in [29]. For the sake of comparison, both been adopted in different power system operation problems, the proposed bilinear cavern model and the nonlinear cavern e.g., transmission congestion relief [31], bidding and offering model in [29] are integrated into the SSP of CAES; both SSPs strategy [32], and unit commitment [33]. An overview of are MINLP problems and can be solved by an MINLP solver, different applications of technologies relevant to CAES is e.g., BARON [46] (note that MIBLP is a kind of MINLP). available in [34]. The constant-temperature model is linear but Furthermore, the SSP of CAES using the proposed bilinear inaccurate, which can result in non-optimal or even infeasible model is an MIBLP model; the MIBLP model is converted into solutions in practice. an MILP model, which is then solved by an MILP solver, e.g., Therefore, all existing CAES applications use simple but CPLEX. inaccurate cavern models. To achieve a balance between In summary, the contributions of this paper include 1) accuracy and complexity, this paper proposes an original and proposing a novel accurate bilinear cavern model of CAES that novel bilinear cavern model. The charging (discharging) is superior to existing linear but inaccurate or accurate but process is divided into four (two) real/virtual states. The ideal highly nonlinear cavern models from an optimization point of gas law and the first law of thermodynamics [35] are used to view, 2) verifying the accuracy of the proposed bilinear cavern model the pressure/temperature relationships between different model via comparison with both an accurate model and field- charging/discharging states. The heat transfer between the measured data, and 3) using the SSP of CAES to verify the cavern air and the cavern wall is also considered. A bilinear superiority of using the bilinear cavern model compared to a cavern model is then obtained by ignoring some very-small- highly nonlinear cavern model in the optimization problem. https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 4 The reminder of this paper is organized as follows. Section Then, let the air in both containers 2 and 3 go into container II details the deduction of the accurate bilinear cavern model. 4. The volume of (mass of air in) container 4 is set to the sum Section III presents the SSP of CAES and flowcharts for using of the volumes of (masses of air in) containers 2 and 3, i.e., the bilinear cavern model. Section IV verifies the effectiveness container 4 can be seen as a combination of containers 2 and 3. of the bilinear cavern model. Section V presents the conclusions Note that the purpose of using virtual containers 2 and 4 is to drawn from the results. let the work be zero during the process of merging the air in containers 2 and 3 into container 4. Last, let the air in container 4 go into container 5, which is an II. ACCURATE BILINEAR CAVERN MODEL FOR COMPRESSED adiabatic compression process as the mass of air does not AIR ENERGY STORAGE change and the volume decreases. The rest of this subsection The cavern of a CAES plant can operate in either constant- details the deduction of the model for the charging process. volume or constant-pressure mode. In this paper, constant- 1) State 1 State 2 volume caverns are considered as they are used in existing The transfer of air from containers 1 to 2 is an adiabatic CAES plants [10]. process and the mass of air does not change. According to the ideal gas law [35] for the air in container 2, one can obtain A. Charging Process = ̇ Δ . (2) In the charging process, a certain amount of air is injected by compressors into a cavern, as shown in Fig. 1. To facilitate the According to the temperature-pressure relationship for an model deduction given in Sections II-A1 to II-A3, the charging adiabatic process, the term / is constant from containers process is divided into four states and the air is assumed to be 1 to 2 [47] and therefore one can obtain stored in the (virtual) containers as shown in Fig. 2, where the five containers are indexed by the numbers in heptagons. The ( ) ( ) characteristics of the air in each container, including the = . (3) pressure, volume, temperature, and mass, are shown in each container in Fig. 2. The values of the underlined notations, i.e., Let represent the left-hand side of (3), i.e., = / . the pressure and temperature in containers 2, 4, and 5, are Note that the constant is known. can be determined unknown while the values of the other notations are known from (1), i.e., = ̇ Δ/ . This leaves only two from the actual CAES operation. unknown variables in (2) and (3), i.e., and . Therefore, Containers 1, 2, and 4 are virtual while containers 3 and 5 and can be obtained from (2) and (3): represent the cavern before and after the air from the compressors is injected into it, respectively. = ( ) / (4) = ( / ) . (5) 2) State 2 State 3 Now we consider the process of the air in both containers 2 and 3 going into container 4. The volume of container 4 is equal Fig. 2. Four states and five containers used for model deduction in the charging to the sum of the volumes of containers 2 and 3. In this process, process. the work is zero and the total internal energy [35] does not change. The change of the internal energy in containers 2 and 3 The air coming out of the compressors is assumed to be is ( − ) and ( − ) , respectively. stored in a virtual container, i.e., cylinder 8 in Fig. 1 and According to the first law of thermodynamics, i.e., = Δ + container 1 in Fig. 2. The conditions of container 1 represent , one can obtain the thermodynamic properties of air at the outlet of the ( ) ( ) − + − = 0. (6) compressors. First, let the air in virtual container 1 go into virtual container 2. The volume of container 2 is set such that Note that can be obtained from (5). is then the only the ratio of the volume of containers 2 to 3 is equal to the ratio unknown variable in (6) and can be expressed as of the mass of air in containers 2 to 3, i.e., ( ) ( ) = + / + . (7) / = ̇ Δ/ (1) According to the ideal gas law for the air in container 4 in where and represent the volumes of containers 2 and 3, Fig. 2, one can obtain respectively; represents the mass of air in container 3; and ( + )= ( + ) . (8) ̇ represents the flow rate of mass charged into the cavern, which is assumed to be constant during a period of time, Δ. By substituting (7) into (8), one can then obtain The mass of air coming out of the compressor during that period ( ) ( ) of time, denoted as , can be expressed as = ̇ Δ. The = = . (9) mass of air injected into the cavern is assumed to be equal to , i.e., there is no air leakage. is also the mass of air in Now and in container 4 have been obtained. both containers 1 and 2. https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 5 3) State 3 State 4 known initial status. When used in a multi-time-step From containers 4 to 5, the mass of air does not change and optimization problem, (17) and (18) are bilinear equations as the volume reduces from ( + ) to . This is an adiabatic and become decision variables. compression process. Thus, / is constant from containers B. Discharging Process 4 to 5 [47], i.e., In the discharging process, the air is released from the cavern, as shown in Fig. 1. To facilitate the model deduction given in ( ) ( ) = (10) the rest of this subsection, the cavern before discharging is divided into two containers, i.e., containers 1 and 2, while the According to the ideal gas law for the air in container 5 in cavern after discharging (cylinder 7 in Fig. 1) is represented as Fig. 2, one has container 3 in Fig. 3. The three containers in Fig. 3 are indexed by the numbers in heptagons. The characteristics of the air in = (̇ Δ + ) (11) each container, including the pressure, volume, temperature, Let represent the left-hand side of (10), i.e., = and mass, are shown in each container in Fig. 3. The values of the underlined notations, i.e., the pressure and temperature in ( ) / . Only two variables are unknown in (10) and container 3, are not known while the values of the other (11), i.e., and . Therefore, and can be obtained notations are known. from (10) and (11): The discharging process can be divided into two virtual steps. ( ) ( ) = ̇ Δ + / (12) First, the air in container 2 is taken out of the cavern. Then, the air in container 1 expands to the whole cavern, i.e., air goes ( ) = . (13) from containers 1 to 3. The deduction of the model for the second step is given in the rest of this subsection. Now and for container 5 have been obtained. By substituting ( , (5), (7), and (9) are needed to calculate ) into (12) and (13), these two equations can be rewritten as: = 1 + + (̇ Δ + ) ̇ Δ (14) Fig. 3. Three containers used for model deduction in the discharging process. ( ) = 1 + + ̇ Δ + ̇ Δ (15) The volume of container 2 is set such that the ratio of the where = and = . volume of container 2 to that of container 1 is equal to the ratio Equations (14) and (15) show that and are nonlinear of the mass of air in container 2 to that of container 1, i.e., functions of ̇ , which are linearized as follows. According to /( − )= /( − ). (19) Newton’s generalized binomial theorem [36], one has Note that the purpose of using virtual container 2 is to let the ( ) (1 + ) = ∑ = 1 + + + ⋯ (16) temperature and pressure of the air in container 1 be the same as the air in the cavern before discharging and to let air go from ( ) ⋯ ( ) containers 1 to 3. This means that step 2 is an adiabatic where = , can be any real number, and is expansion process. an integer. That is, 1 + in (14) can be expressed as Let ̇ represent the flow rate of mass discharged from the cavern, which is assumed to be constant during a period of time, ̇ ( )( ) ̇ 1 + ( − 1) + + ⋯ . Considering Δ. Then, the mass of air discharged from the cavern, denoted as , during that period of time can be expressed as = that ̇ Δ is much smaller than , the second and higher ̇ Δ. orders terms of (̇ Δ/ ) can be ignored. Then, (14) can be The air expansion from containers 1 to 3 is an adiabatic reformed as process. Then, / is constant from containers 1 to 3: ( ) = 1 + ( − 1) + ̇ Δ (17) ( ) = ( ) . (20) Note that the second term in (14) is replaced by ( ) ̇ Δ, which has negligible error as is much According to the ideal gas law for the air in container 3 in smaller than (e.g., =46~66× 10 Pascals and =1.04× Fig. 3, one has 10 for the Huntorf CAES plant) and ̇ Δ is much smaller = ( − ) . (21) than . Similarly, (15) can be reformed as There are only two variables unknown in (20) and (21), i.e., ( ) = 1 + ( − 2) + ̇ Δ (18) and . Therefore, and can be obtained from (20) and (21): When (17) and (18) are used in a single-time-step = (1 − ̇ Δ/ ) (22) optimization problem, is a linear function of ̇ in (17) and is a linear function of ̇ in (18) as and have a = (1 − ̇ Δ/ ) . (23) https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 6 Similar to the charging process, the above nonlinear where superscript ‘a,ht’ indicates that both the adiabatic process equations are linearized. According to Newton’s generalized and heat transfer are considered. binomial theorem, when ̇ Δ is much smaller than , (22) According to the first-order Taylor series approximation [36], ( ) ( ) ( ) and (23) can be respectively reformulated as i.e., = + × ( − ) , one can linearize ( ) ( ) and at as = (1 − ̇ Δ/ ) (24) ( ) ( ) ( )( ) = + − 1 ( − ) (33) = (1 − ( − 1)̇ Δ/ ) . (25) ( ) ( ) ( )( ) = + − 2 ( − ) (34) Similar to (17) and (18), (24) and (25) are linear (bilinear) equations when used in a one-step (multi-step) optimization where is a fixed value, i.e., = . The ̇ problem. and ̇ in (32) can be linearized using . . C. Charging Process Considering Heat Transfer ( ) ̇ = ̇ + ̇ − ̇ (35) In Sections II-A and II-B, the heat transfer between the . . ̇ = ̇ + (̇ − ̇ ) (36) cavern air and the cavern wall is not considered. However, the . . . . ̇ ̇ ̇ ̇ heat transfer plays an important role in the variation of the air where = , = , and the setting ̇ ̇ ̇ ̇ temperature/pressure in the cavern [28]. Therefore, the heat of ̇ and ̇ will be given in the beginning of Section IV. transfer is considered in this subsection and the following two By using (33)-(36), equation (32) can then be converted to be a subsections. In this subsection, the temperature as a function of bilinear form, i.e., (37). time is first deduced. The pressure as a function of time is then obtained via the ideal gas law. Last, the temperature/pressure as ( ) = + ̇ + ̇ + c ̇ + + a function of time is linearized to obtain a bilinear model. c + (37) According to [28], [29], the air density ( ) in the cavern and the cavern wall temperature ( ) can be assumed to be where - are constant coefficients given in the Appendix. constant and the heat transfer between the cavern air and the ( ) Equation (37) involves four variables, i.e., , ̇ , , cavern wall can be modelled as and and the first four terms are bilinear terms. Equation (37) represents the change of the temperature = ( − ) (26) during the charging process as a function of time and the charging mass flow rate ̇ , where both the adiabatic process where and the heat transfer process are considered. ℎ = ℎ + ℎ |̇ − ̇ | (27) According to the ideal gas law, one can obtain , , According to [28], ℎ and ℎ can be set to 0.2356 and 0.0149, ( ) ( ) = + ̇ ()/ (38) respectively. Equation (26) can be reformed as which can be expanded to (39) by substituting (32) therein. ()= ( − ) (28) ∫ ( ) ̇ ̇ ( ) ( ) = 1 + − 2 ( ̇ ) Equation (18) can be written as ( ) + ̇ ( ̇ ) ̇ ()= 1 + ( − 2) + ( ) ̇ . (29) + ℎ + ℎ ̇ − − ( − 2) ( ) By substituting (29) into (28), i.e., replacing on the right- ( ) − ℎ + ℎ ̇ ̇ (39) hand side of (28) by the right-hand side of (29), one can obtain The four terms in (39) are each on a separate line. The second ( ) ( ) = ∫ − 1 + − 2 − term of (39) can be replaced by ( ) ̇ / because / is small and ̇ is much smaller than . ( ) ̇ (30) The last term of (39) can be ignored because the values of both ℎ + ℎ ̇ / and / are small. where superscript ‘ht’ represents ‘heat transfer’. By solving the Again, according to the first-order Taylor series integral equation (30), one can obtain approximation [36], i.e., ()= ( )+ ( )× ( − ), ( ) ( ) = − + − 2 − one can linearize at : ( ) ( ) ̇ /2. (31) = + − (40) Now, by using (40), (39) can be reformed into a bilinear Adding (29) and (31) together gives form: ()= 1 + ( − 2) + ( ) ̇ + ()= + ̇ + c ̇ + ̇ + ℎ + ℎ ̇ − + ( − 2) − + + (41) ( ) ̇ (32) https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 7 ( ̇ ) ( ̇ ) ̇ where - are constant coefficients as given in the Appendix. ()= − ( − 1) + ( ) Equation (41) involves four variables, i.e., , ̇ , , ( ) ( ) ̇ ̇ ( ) ( − + − and , and the first four terms are bilinear terms. 1) (51) D. Discharging Process Considering Heat Transfer In this subsection, the temperature as a function of time is ̇ ( ) Note that − ̇ in the third term of (51) can be first deduced. The pressure as a function of time is then obtained via the ideal gas law. Last, the temperature/pressure as a replaced by because ̇ is much smaller than . function of time is linearized to obtain a bilinear model. Equation (51) can then be reformed as Equation (25) can be written as ()= ( − ̇ ) ()= − ( − 1) . (42) + (( − )( − ) + 0.5( − 1) ̇ ) By substituting (42) into (28), i.e., replacing on the right- (52) hand side of (28) by the right-hand side of (42), one can obtain Comparing (52) with (24), we know that the term ()= ∫ − + ( − 1) d. (43) ( − ̇ ) in (52) represents the adiabatic process inside the cavern and the terms in the second line of (52) are By solving the integral equation (43), one can obtain associated with the heat transfer between the cavern air and the cavern wall. Equation (52) can be further reformulated into the ()= ( − )+ ( − 1) . (44) following bilinear form: ()= + ̇ + + ̇ + Adding (42) and (44) together gives ̇ ̇ + ̇ + ̇ + + + ()= − ( − 1) + ( − ) + ( (53) ̇ where - are constant coefficients as given in the − 1) . (45) Appendix. Equation (53) involves five variables, i.e., , ̇ , ( ) , , and and the first seven terms are bilinear terms. Equation (45) represents the change in temperature during the discharging process as a function of time and discharging E. Idle Process Considering Heat Transfer mass flow rate ̇ , where both the adiabatic process and the When idling, i.e., when neither charging nor discharging, heat transfer process are considered. Equation (45) can be heat transfer occurs between the cavern air and the cavern wall rewritten as if a temperature difference exists. By solving the integral equation (28), the change of air temperature in the cavern in the ( ) ()= − ( − 1) ̇ + − idle process can be expressed as . . ℎ + ℎ ̇ + ( − 1) ℎ ̇ + ℎ ̇ (46) ()= ( − ) + (54) . . Similarly, ̇ and ̇ can be linearized using where is the initial air temperature of the cavern in the idle process. According to the ideal gas law, one can obtain . . ̇ = ̇ + (̇ − ̇ ) (47) ( ) = ()/ (55) . . ( ) ̇ = ̇ + ̇ − ̇ (48) which can be expanded into (56) by substituting (54) therein: . . . . ̇ ̇ ̇ ̇ where = , = , and the ̇ ̇ ̇ ̇ ()= ( − ) / + / (56) setting of ̇ and ̇ will be given at the beginning of Section IV. Now, (46) can be reformulated into a bilinear form: Equation (56) can be reformed as ()= + ̇ + c ̇ + + ( ) = + (1 − )/ (57) (49) where - are constant coefficients as given in the The in (57) can be expressed as , which Appendix. Equation (49) involves four variables, i.e., , ̇ , can be linearized as follows: (), and , and the first three terms in the equation are bilinear terms. According to the ideal gas law, one can obtain = + ( − ) (58) , , ()= ( − ̇ ) ()/ (50) where = . By substituting (58) into (54) and (57), which can be expanded as follows by substituting (45) therein: one can obtain https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 8 ≤ ≤ , ∀ ∈ Ω (66) ()= ( − )( + ( − )/ )+ (59) + ≤ 1, ∀ ∈ Ω (67) ()= ( + ( − )/ ) ≤ ≤ , ∀ ∈ Ω (68) + 1 − − ( − ) (60) ≤ ≤ , ∀ ∈ Ω (69) Equations (59) and (60) can be reformulated into the following bilinear forms: where the values of the coefficients and are adopted from [26]. ( ) = + + + (61) The bilinear cavern model presented in the previous section, ()= + + + (62) i.e., (37), (41), (49), (53), (61) and (62), is re-written below where - are constant coefficients as given in the using superscript and (+1) to represent the indices of time Appendix. Equation (61) involves three variables, i.e., , steps. nd (), and , and the 2 term is the only bilinear term. ( ) ( ) ( ) ( ) Equation (62) involves three variables, i.e., , (), and , = + ̇ + ̇ + c ̇ + nd rd and the 2 and 3 terms are bilinear. + c + , ∀ ∈ Ω (70) This completes the deduction of the bilinear cavern model. In summary, the bilinear cavern model includes (37) and (41) ( ) ( ) ( ) ( ) = + ̇ + c ̇ + ̇ + for the charging process, (49) and (53) for the discharging , process, and (61) and (62) for the idle process. + + , ∀ ∈ Ω (71) ( ) ( ) ( ) = + ̇ + c ̇ + + , III. SELF-SCHEDULING OF COMPRESSED AIR ENERGY ∀ ∈ Ω (72) STORAGE IN THE DAY-AHEAD ELECTRICITY MARKET ( ) ( ) ( ) A. Self-scheduling Problem Using the Bilinear Cavern Model = + ̇ + + ̇ + ( ) ( ) ( ) ( ) In the SSP of CAES, the day-ahead electricity market is ̇ ̇ + ̇ + ̇ + + considered and the CAES plant is assumed to be a price taker + , ∀ ∈ Ω (73) (i.e., the CAES plant does not affect the electricity market price). ( ) The objective of SSP is to maximize the arbitrage revenue = + + + , ∀ ∈ Ω (74) obtained from selling electricity to and buying electricity from ( ) the day-ahead electricity market, i.e., selling electricity (in the = + + + , ∀ ∈ Ω discharging process) to the market in high-electricity-price (75) periods and buying electricity (in the charging process) from ( ) ( ) ( ) the market in lower-electricity-price periods. This implies that The ( , ) represents the temperature of the , , , the aim of SSP is to decide when to charge and discharge cavern air if the th step is a charging (discharging, idle) process. subjected to physical constraints of the CAES plant, including ( ) The temperature at the beginning of the (+1)th step, i.e., , the minimum/maximum pressure range of the cavern air. ( ) ( ) ( ) is equal to one of , , and according to the The objective function of the SSP can be expressed by (63), , , , ( ) ( ) which represents the net profit obtained by the CAES plant values of and , which is modelled in (76). The from the electricity market. Note that the operational costs of relationship between the pressure at the beginning of the ( ) ( ) ( ) ( ) charging and discharging power in the CAES are considered in (+1)th step, i.e., , and , , and can be , , , (63). Equation (64) represents the relationship between the similarly modeled as (77). The relationship among the mass of mass flow rate in (̇ ) and charging power ( ) and (65) ( ) air at two consecutive steps ( and ) and the represents the relationship between the mass flow rate out (̇ ) ( ) ( ) charging/discharging status indicators ( and ) can and discharging power ( ) [12]. Equation (66) represents the be expressed as (78). lower and upper bounds of the air pressure in the cavern. Equation (67) ensures that the charging and discharging ( ) ( ) ( ) ( ) ( ) ( ) = + + 1 − − , , processes do not occur at the same time. Equations (68) and (69) ( ) ( ) describe the ranges of charging and discharging power and their , ∀ ∈ Ω (76) relationships with the charging and discharging status ( ) ( ) ( ) ( ) ( ) ( ) indicators, i.e., and , respectively. = + + (1 − − , , ( ) ( ) maximize: = ∑ ( − ) − ( + ∈Ω ) , ∀ ∈ Ω (77) ) (63) ( ) ( ) ( ) ( ) ( ) = + ̇ Δ − ̇ Δ, ∀ ∈ ̇ = , ∀ ∈ Ω (64) Ω (78) The SSP of CAES using the bilinear model, i.e., (63)-(78), is ̇ = , ∀ ∈ Ω (65) https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 9 referred to as Model 1, which is an MIBLP model. The bilinear terms in Model 1 can be expressed as , which is equal to ( ) ( ) ( ) ( ) + /4 − − /4 . Then, + and − can be piecewise linearized. The detailed procedure of this linearization is available in Section IV-C of [40]. The resulting MILP model is referred to as Model 2. B. Flowcharts for Using the Bilinear Cavern Model The flowcharts for using the bilinear model, i.e., (70)-(78), to calculate the temperature, pressure, and mass of air in the charging, discharging, and idle processes are given in Figs. 4a, 4b, and 4c, respectively. The calculation procedures given in Figs. 4a, 4b, and 4c are recursive but consume very little time because equations (70)-(75) each only have one variable per step that can simply be calculated from the other known values. For example, is the only variable in (70) and all the other notations in (70) are known. Flowcharts given in Figs. 4a, 4b, (c) (d) and 4c are used in Sections IV-A to IV-C. Fig. 4. Flowchart for calculating the temperature, pressure, and mass of air using the bilinear model: (a) in the charging process, (b) in the discharging The flowchart for using the bilinear cavern model for an process, (c) in the idle process; (d) flowchart for using the proposed bilinear optimization problem is given in Fig. 4d, i.e., using (70)-(78) in model in an optimization model. an optimization problem. The bilinear cavern model part, i.e., (70)-(78), in an optimization model has 11 continuous C. Self-scheduling Problem Using Nonlinear and Constant- variables (i.e., , , and for = 1,2, ⋯ , and , temperature Cavern Models , , , , , ̇ , and ̇ for = , , , , , For the sake of completeness, the nonlinear cavern model 0,1,2, ⋯ , ( − 1)) and 2 binary variables (i.e., and given in [29] is rewritten here, i.e., (79)-(82). The SSP of CAES for = 0,1,2, ⋯ , ( − 1)). The flowchart given in Fig. using the nonlinear cavern model given in [29] consists of (63)- 4d is used in Sections IV-D and IV-E. Equations (70)-(78) (69), (76), and (78)-(82), and is referred to as Model 3. together with the objective and other constraints, e.g., (63)-(69), ( ) ( ) form an optimization problem that can be solved by an MINLP ( ) = + − ( ) solver, further details of which are given in Section IV-D. ( ) (79) ( ) Input , , , and at Input , , , and at ( ) time step time step ( ) = + − , ( ) Calculate and via Calculate and via (80) ( ) (70) and (71), respectively (72) and (73), respectively , , and ( + ) , , and ( ) ( ) = ( − ) + (81) are respectively used as , are respectively used as , , and for the next , and for the next ( ) ( ) ( ) = / (82) step in the charging process step in the charging process For the sake of comparison, the constant-temperature cavern model is also provided here, i.e., (83). The SSP of CAES using the constant-temperature model consists of (63)-(69), (76), and end ? end ? (83), which is referred to as Model 4. Models 1-4 will be compared in Section IV-E. (a) (b) ( ) = (83) IV. SIMULATION In this paper, the parameters for the Huntorf CAES plant are used for the calculations. The Huntorf CAES plant features two caverns with volumes of 141,000 and 169,000 m , respectively. Note that the maximum mass flow rate in the charging process (̇ ) is 108 kg/s for the whole plant and 49.12 kg/s for the first cavern, which is calculated from 108× . Similarly, the maximum mass flow rate in the discharging process (̇ ) https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 10 is 417 kg/s for the whole plant and 189.67 kg/s for the first the inaccuracy of the charging/discharging parts of the bilinear model, i.e., (70)-(73), is around 0.11%, which is quite small. cavern, which is calculated from 417× . In this This verifies the accuracy of the bilinear model for the given paper, the first cavern is used for calculations. The ̇ and initial temperature/pressure and mass flow rate. ̇ used in (35) and (36) are set to 49.12 and 22.74 kg/s, respectively. The ̇ and ̇ used in (47) and (48) are set TABLE II to 189.67 and 90.97 kg/s, respectively. The temperature at the THE MAPE BETWEEN THE RESULTS OBTAINED BY THE BILINEAR MODEL AND outlet of high-pressure compressor is high (about 230 °C), but THE ANALYTICAL MODEL GIVEN IN [29] IN EACH OF THE THREE PROCESSES. Charging process Discharging process Idle process drops to 50 °C after going through an aftercooler that is located Pressure 0.0011 0.0011 1.12× 10 between the high-pressure compressor and the cavern, i.e., the Temperature 0.0011 0.0011 1.12× 10 temperature of air injected into the caverns is equal to 50 °C. The other parameters for the Huntorf CAES plant are given in 70 50 Table I [28], [29]. TABLE I PARAMETERS FOR THE HUNTORF CAES PLANT. 25,000 m 718.3 J/(kg K) 1.4 66 bar 3 \$/MWh 286.7 J/(kg K) 141,000 m 40 °C 50 °C 3 \$/MWh Bilinear model Bilinear model Analytical model Analytical model Three specific processes are defined as follows and used to 45 20 0 4 8 12 16 0 4 8 12 16 verify the accuracy of the proposed bilinear model: Hour Hour  Charging process: Set the initial pressure (temperature) of (a) (b) the air in the cavern to 46 bar (20 °C). Charge the first Fig. 5. Results obtained by the proposed bilinear model and the analytical model cavern (141,000 m ) continuously for 16 hours at the in [29] during the charging process: a) pressure, b) temperature. maximum mass flow rate, i.e., ̇ =49.12 kg/s. 70 40  Discharging process: Set the initial pressure (temperature) Bilinear model Bilinear model Analytical model Analytical model of the air in the cavern to 66 bar (40 °C). Discharge the cavern continuously for 4 hours at the maximum mass flow rate, i.e., ̇ =189.67 kg/s.  Idle process: Set the initial pressure (temperature) of the air in the cavern to 60 bar (45 °C). Let the cavern be in the idle process for 16 hours. Reference [29] compares several existing CAES models with the measured data from the Huntorf CAES plant. The analytical 45 20 0 1 2 3 4 0 1 2 3 4 model in [29] is accurate and simpler than other existing Hour Hour analytical models. Thus, the analytical model in [29] is used as (a) (b) a benchmark model in this section to verify the accuracy of the Fig. 6. Results obtained by the proposed bilinear model and the analytical model proposed bilinear cavern model. in [29] during the discharging process: a) pressure, b) temperature. A. Model Verification - Comparison With Accurate Model In this subsection, the time interval is set to 1 second, i.e., Δ is equal to 1 second in (70)-(75). The pressure and temperature for each time interval of the charging (discharging, idle) process obtained from both the proposed bilinear model and the analytical model in [29] are plotted in Fig. 5 (Fig. 6, Fig. 7). Note that the difference between the results of the two models would be difficult to observe if the results for each second were shown in Figs. 5-7; therefore, the results for every 1000 (250, 1000) seconds are shown in Fig. 5 (Fig. 6, Fig. 7). Figs. 5-7 show that the pressure/temperature results obtained from the (a) (b) proposed bilinear model and the analytical model are quite Fig. 7. Results obtained by the proposed bilinear model and the analytical model close to one another. in [29] during the idle process: a) pressure, b) temperature. The mean absolute percentage errors (MAPEs) between the To comprehensively evaluate the accuracy of the bilinear results, in terms of both pressure and temperature, obtained model, different settings of initial temperatures/pressures and from the bilinear model and the analytical model during the mass flow rates are used in the charging/discharging/idle charging, discharging, and idle processes are tabulated in Table process. Different settings for the charging (discharging, idle) II. The last column of Table II shows that the idle part of the process are shown in rows C1-C6 (D1-D7, I1-I4) of Table III. bilinear model, i.e., (74) and (75), is almost as accurate as the nd rd When the pressure is larger (smaller) than 46 bar, the CAES is analytical model. The 2 and 3 columns of Table II show that Pressure (bar) Temperature ( C) https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 11 in normal (emergency) operating mode. Fig. 5 (Fig. 6) shows For each setting, the MAPE and mean absolute error (MAE) that the cavern temperature is relatively high (low) when the between the results (including both pressure and temperature) pressure is high (low). Therefore, the initial temperatures are set obtained by the bilinear and the analytical models is given in to 20 and 35 °C (50 and 35 °C) in the different settings for the the last four columns of Table III. The largest error in different charging (discharging) process as its initial pressure is settings for the charging process occurs at C2 with low initial relatively low (high). Both large and small mass flow rates are temperature and low mass flow rate; the MAPE and MAE for investigated in the normal operating mode as given in C1-C4 pressure (temperature) are 0.77% and 0.38 bar (0.77% and and D1-D4 where the large and small values are set to the 2.38 °C), respectively. The largest error in different settings for maximum mass flow rate and one tenth of it, respectively. To the discharging process occurs at D5 with high initial avoid a lengthy table, 1) only results that tend to have a large temperature and low mass flow rate; the MAPE and MAE for error in the emergency model are listed in Table III, i.e., low pressure (temperature) are 0.82% and 0.36 bar (0.82% and mass flow rate (associated with low charging power) and low 2.58 °C), respectively. Note that the MAPE and MAE are initial temperature as indicated by setting C1-C4, and low mass usually less than the worst-case values given above. The errors flow rate (associated with low discharging power) and high in different settings for the idle process are very small; the initial temperature as indicated by settings D1-D4; and 2) not MAPE for both pressure and temperature is less than 5E-5. This all the results of different settings for the idle process are listed indicates that the error between the proposed bilinear cavern because all of the different settings have very small errors, i.e., model and the accurate analytical model is small under a variety the MAPE is less than 5E-5. Note that ‘E-n’ and ‘E+n’ are used of circumstances. in this paper to represent ‘× 10 ’ and ‘× 10 ’, respectively. TABLE III DIFFERENT SETTINGS OF INITIAL PRESSURES AND TEMPERATURES AND CHARGING/DISCHARGING POWER AND THE CORRESPONDING MAPE AND MAE BETWEEN RESULTS OBTAINED BY THE BILINEAR CAVERN MODEL AND THE ANALYTICAL MODEL GIVEN IN [29] IN EACH OF THE THREE PROCESSES Pressure Temperature Setting Initial pressure (bar) Initial temperature (°C) Mass flow rate (kg/s) MAPE MAE (bar) MAPE MAE (°C) C1 46 20 49.12 0.0011 0.07 0.0011 0.35 C2 46 20 4.912 0.0077 0.38 0.0077 2.38 C3 46 35 49.12 0.0021 0.11 0.0020 0.64 C4 46 35 4.912 0.0017 0.08 0.0017 0.53 C5 30 20 4.912 0.0060 0.19 0.0060 1.84 C6 5 20 4.912 0.0047 0.03 0.0047 1.46 D1 66 50 189.67 0.0018 0.10 0.0018 0.57 D2 66 50 18.967 0.0078 0.50 0.0078 2.47 D3 66 35 189.67 0.0008 0.04 0.0008 0.24 D4 66 35 18.967 0.0056 0.37 0.0056 1.75 D5 46 50 18.967 0.0082 0.36 0.0082 2.58 D6 30 50 18.967 0.0071 0.20 0.0071 2.22 D7 5 50 18.967 0.012 0.047 0.012 3.82 I1 46 20 ---- 3.9E-5 1.8E-3 3.9E-5 1.2E-2 I2 5 20 ---- 4.2E-6 2.2E-5 4.2E-6 1.3E-3 I3 66 50 ---- 2.4E-5 1.6E-3 2.4E-5 7.7E-3 I4 5 50 ---- 1.8E-6 9.1E-6 1.8E-6 5.9E-4 Kelvin when calculating the MAPE (Kelvin is the unit of B. Model Verification - Comparison With Field-Measured temperature for all calculations involved in the bilinear cavern Data model). Note that the cavern pressure being higher (lower) than To further verify the accuracy of the proposed bilinear model, 46 bar indicates normal (emergency) operating mode; the CAES usually runs in normal operating mode. Therefore, the field-measured pressure and temperature data during a discharging process from [28] and a 10-stage proposed bilinear model is quite accurate when compared to field-measured data I in the normal operating mode of CAES. charging/discharging/idle process from [29] are used and referred to as field-measured data I and II, respectively. The For field-measured data II, the air mass flow rates in the 10 stages are given in Fig. 9a where negative (positive) values pressure and temperature results from both the proposed bilinear model and field-measured data I are plotted in Figs. 8b represent the discharging (charging) process and the zero value represents the idle process. In Fig. 9a, the 10 stages are divided and 8c, respectively. These figures show the pressures/temperatures obtained from the proposed bilinear by nine vertical dashed lines. In stage 1, the air mass flow rate linearly decreases from -125 to -175 kg/s and then increases to model are close to field-measured data I. Figs. 8b and 8c are both divided by a dashed line into two parts: the left-hand -125 kg/s. In stage 5, the CAES is discharging at a fixed air mass flow rate of -175 kg/s. In stages 3, 7, and 9, the CAES is (right-hand) part corresponds to pressure being higher (lower) than 46 bar. For the left-hand (right-hand) part, the MAPE and charging at a fixed air mass flow rate of 50 kg/s. In stages 2, 4, 6, 8, and 10, the CAES is in the idle stage, i.e., the air mass flow MAE between the results obtained by the bilinear model and field-measured data I are 0.47% and 0.27 bar (5.85% and 1.94 rate is zero. The pressures (temperatures) at each stage obtained from the bilinear model, analytical model, and field-measured bar) for the pressure and 0.19% and 0.56 °C (0.23% and 0.64 °C) for the temperature, where the unit of temperature is set to data II are plotted in Fig. 9b (Fig. 9c). Fig. 9b shows the https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 12 pressures obtained from both the bilinear model and the analytical model are quite close to field-measured data II except for stage 2. The MAPE and MAE between the pressure obtained by the bilinear (analytical) model and field-measured data II are (b) 1.01% and 0.54 bar (0.97% and 0.51 bar), respectively. In Fig. 9c, the temperature obtained from the bilinear model is closer to field-measured data II than the analytical model. The MAPE and MAE between the temperature obtained by the bilinear (analytical) model and field-measured data II are 3.16% and 1.18 °C (3.88% and 1.47 °C), respectively. In summary, the pressure (temperature) result obtained from the bilinear cavern model can well match the pressure (temperature) from both field-measured data I and II with a relative error of 0.47 and 1.01% (0.19 and 3.16%), respectively, in the normal operating mode of CAES. This comparison with (c) both an accurate model and two sets of field-measured data verifies the accuracy of the proposed bilinear cavern model. Bilinear model Analytical model Field-measured data II 0 5 10 15 20 25 Hour (a) Fig. 9. Mass flow rate into cavern (a) and comparison between the results obtained by the bilinear cavern model, analytical cavern model, and field- measured data II obtained from [29] for pressure (b) and temperature (c). C. Impact of Heat Transfer and Temperature To observe the impact of heat transfer, the results obtained (b) from the bilinear model with and without considering heat transfer in the charging (discharging) process are plotted in Fig. 10 (Fig. 11). In Fig. 10b, the slope of the left-hand part and the right-hand part of the dotted line (results of the model with heat transfer) is higher and lower, respectively, than the circled line (results of the model without heat transfer). The reason is that (c) when the cavern air temperature is lower (higher) than the cavern wall temperature, i.e., 40 °C, the heat transfer provides heat to (absorbs heat from) the cavern air and therefore the temperature obtained by the model with heat transfer increases faster (slower) than the model without heat transfer. According to the ideal gas law, i.e., = , for the same , , and , Fig. 8. Mass flow rate out of cavern (a) and comparison between the results higher (lower) temperature corresponds to higher (lower) obtained by the bilinear cavern model and field-measured data I from [28] for pressure (b) and temperature (c). pressure and therefore the relationship between the dotted and circled lines in Fig. 10a is similar to Fig. 10b. In Fig. 11, the dotted line has a smaller slope than the circled line. The reason is that the heat transfer provides heat to the cavern air and therefore the temperature decrease associated with the model with heat transfer is smaller than the model without heat transfer. Therefore, heat transfer clearly has a significant impact (a) on the temperature and pressure variation and, consequently, it is important to consider heat transfer in the cavern model. The results obtained from the constant-temperature model in the charging (discharging) process are also plotted in Fig. 10 (Fig. 11). Obviously, the pressure and temperature obtained from the constant-temperature model are quite different from those obtained with the bilinear model. Considering that the accuracy of the bilinear model was verified in Section V-A, Figs. 10 and 11 indicate that the constant-temperature cavern model is inaccurate. Therefore, it is necessary to use an accurate cavern model instead of the constant-temperature cavern model, especially in a real application problem. Pressure (bar) https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 13 Figs. 12a-12d show that the accuracy of the bilinear model in the charging and discharging processes decreases as the time interval increases. This is because the charging and discharging parts of the model utilize the assumption that the air injected into or released from the cavern in each time step is much smaller than the total amount of air in the cavern (i.e., ̇ Δ ≪ and ̇ Δ ≪ ); this assumption introduces higher error when the time interval is larger. Figs. 12e-12f show that the accuracy of the bilinear model in the idle process does not change with the time interval; this is because the idle part of the model does not utilize the above assumption. Tables IV-V show that the error and relative error of the temperature and pressure in both the charging and discharging processes are small when the time interval is less than or equal to 10 minutes. When the time interval is equal to 20 minutes, (a) (b) the errors (relative errors) of the pressure in the charging and Fig. 10. Results obtained from the bilinear cavern model (with and without discharging processes are 0.121 and 0.045 bar (0.17 and 0.09%), considering heat transfer) and the constant-temperature cavern model in the respectively, which are still relatively small. When the time charging process: a) pressure of the air in the cavern and b) temperature of the air in the cavern. interval is equal to 60 minutes, the error (relative error) in both the charging and discharging processes is relatively large. Table VI shows that the idle process of the model is quite accurate for all time intervals. Therefore, Fig. 12 and Tables IV-VI show that the accuracy of the bilinear cavern model is high, moderate, and relatively low when the time interval is between 1 second and 10 minutes, equal to 20 minutes, and equal to 60 minutes, respectively. (a) (b) Fig. 11. Results obtained from the bilinear cavern model (with and without considering heat transfer) and the constant-temperature cavern model in the discharging process: (a) pressure of the air in the cavern and (b) temperature of (a) (b) the air in the cavern. D. Different Time Intervals In power system operation problems, the time interval is usually longer than one second, e.g., the time intervals of economic dispatch and unit commitment are both usually 1 hour, respectively. Therefore, it is necessary to know whether the proposed bilinear model is accurate for different time intervals. (c) (d) In this regard, the status of the air in the cavern is calculated using the procedure given in Figs. 4a, 4b, and 4c for different time intervals (1 second, 1 minute, 5 minutes, 10 minutes, 20 minutes, and 60 minutes) using the same initial status. The final temperature and pressure of the charging, discharging, and idle processes obtained by the bilinear model and the analytical model are plotted in Fig. 12. The error and relative error (values given in round brackets) between the results, in terms of final temperature and pressure, obtained from the two models are (e) (f) nd th th th th shown in Table IV, where the 2 -5 , the 6 -9 , and the 10 - Fig. 12. Final temperature/pressure obtained by both the analytical model and th 13 rows show the error/relative error in the charging, the bilinear model in different processes using different time intervals: (a) final discharging, and idle processes, respectively. Again, the unit of temperature in the charging process, (b) final pressure in the charging process, (c) final temperature in the discharging process, (d) final pressure in the temperature is set to Kelvin when calculating the relative error discharging process, (e) final temperature in the idle process, and (f) final of temperature. pressure in the idle process. Pressure (bar) Pressure (bar) o Temperature ( C) Temperature ( C) o o o Temperature ( C) Temperature ( C) Temperature ( C) Pressure (bar) Pressure (bar) Pressure (bar) https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 14 TABLE IV value represents a better solution as the goal of SSP is to ERROR (RELATIVE ERROR IN ROUND BRACKETS) BETWEEN THE SOLUTION maximize the profits. When the numbers of time steps are 10, OBTAINED BY THE BILINEAR MODEL AND THE ANALYTICAL MODEL IN THE 24, and 72, the objectives of Model 2 are 1.3, 7.4, and 3.6% CHARGING PROCESS USING DIFFERENT TIME INTERVALS higher than Model 3 and 1.3, 7.4, and 1.1% higher than Model Interval 1 s 1 min 5 min 10 min 20 min 60 min 1, respectively. 0.066 0.051 -0.010 -0.087 -0.241 -0.872 Temperat The electricity market price, power output, and pressure in ure (°C) (0.02%) (0.02%) (-3E-5) (-0.03%) (-0.08%) (-0.27%) the 24-time-step case are plotted in Fig. 13. Note that the power 0.002 0.007 0.031 0.061 0.121 0.356 Pressure( output of the CAES in Fig. 13b equals the discharging power bar) (0.002%) (0.01%) (0.05%) (0.09%) (0.17%) (0.52%) minus the charging power. The high price hours are 9-12 and 18-21. Fig. 13b indicates that the CAES generally discharges TABLE V power at high price hours and charges power at low price hours ERROR (RELATIVE ERROR IN ROUND BRACKETS) BETWEEN THE SOLUTION to maximize the profit. The difference between the results of OBTAINED BY THE BILINEAR MODEL AND THE ANALYTICAL MODEL IN THE Models 1-3 lies in the power output in hours 1, 12, 18, and 20- DISCHARGING PROCESS USING DIFFERENT TIME INTERVALS 22. A detailed comparison between Models 2 and 3 is given Interval 1 s 1 min 5 min 10 min 20 min 60 min below. A similar analysis can be applied to Models 2 and 1 but -0.386 -0.350 -0.201 -0.014 0.366 1.968 Temperat is not provided. ure (°C) (-0.13%) (-0.12%) (-0.07%) (-5E-5) (0.12%) (0.67%) The results of Model 2 discharge more power at hours 12 and -0.060 -0.055 -0.034 -0.008 0.045 0.270 Pressure 21 and charge more power at hour 20 compared to the results (bar) (-0.13%) (-0.12%) (-0.07%) (-0.02%) (0.09%) (0.58%) of Model 3, as detailed in Table VIII. In Table VIII, the last column provides the total profit in hours 12, 20, and 21 and TABLE VI clearly shows that Model 2 achieves higher profits than Model ERROR (RELATIVE ERROR IN ROUND BRACKETS) BETWEEN THE SOLUTION 3 and coincides with Table VII. OBTAINED BY THE BILINEAR MODEL AND THE ANALYTICAL MODEL IN THE IDLE PROCESS USING DIFFERENT TIME INTERVALS Linearizing the bilinear model into an MILP problem results Interval 1 s 1 min 5 min 10 min 20 min 60 min in high accuracy as shown in [40], i.e., Model 2 is almost as -2E-4 -2E-4 -2E-4 -2E-4 -2E-4 -2E-4 accurate as Model 1. Therefore, the higher profit obtained from Temperat ure (°C) Model 2 indicates that a better solution has been obtained from (-1E-6) (-1E-6) (-1E-6) (-1E-6) (-1E-6) (-1E-6) Model 2 than Models 1 and 3 and the solutions obtained from -1E-4 -1E-4 -1E-4 -1E-4 -1E-4 -1E-4 Pressure( Models 1 and 3 are not optimal. This is because MILP problems bar) (-2E-6) (-2E-6) (-2E-6) (-2E-6) (-2E-6) (-2E-6) are easier to solve than MINLP or MIBLP problems. This shows the advantage of the proposed bilinear cavern model, E. Self-scheduling Problem of Compressed Air Energy which can be converted to an MILP problem and is therefore Storage in Day-ahead Electricity Market suitable for integration into optimization problems considering This subsection presents the results of solving the SSP of CAES. CAES using different cavern models. When the time interval is set to 60 minutes, the day-ahead SSP has 24 steps, which TABLE VII introduces a relatively small computational burden. However, OBJECTIVE VALUES (\$) OF THE SELF-SCHEDULING PROBLEM OF COMPRESSED the error is relatively large as shown in the previous subsection. AIR ENERGY STORAGE USING DIFFERENT CAVERN MODELS (NUMBERS IN When the time interval is set to 20 minutes, the error is BRACKETS ARE THE RELATIVE DIFFERENCE BETWEEN MODEL 2 AND MODEL (= 1,3)) relatively small and the day-ahead SSP has 72 steps. When the Number of time Model 3 - Model 1 - Model 2 - time interval is set to 10 minutes or smaller, the error decreases steps ( ) MINLP [29] MIBLP MILP but the computational burden increases. Therefore, the time 10 2.28E+5 (1.3%) 2.28E+5 (1.3%) 2.31E+5 interval in the SSP is set to 20 minutes as a trade-off between 24 2.51E+5 (7.4%) 2.51E+5 (7.4%) 2.71E+5 72 5.43E+5 (3.6%) 5.57E+5 (1.1%) 5.63E+5 the accuracy and computational burden. 1) Comparison Among Models 1-3 Models 1 and 3, described in Sections III-A and III-C, TABLE VIII respectively, are both solved by a popularly used MINLP solver, THE PRICE AND THE SOLUTION OBTAINED BY MODELS 2 AND 3 IN HOURS 12, 20, AND 21 OF THE SELF-SCHEDULING PROBLEM OF COMPRESSED AIR i.e., BARON [46], with the values of objective functions (i.e., ENERGY STORAGE rd nd the profits from the electricity market) given in the 3 and 2 Hour 12 Hour 20 Hour 21 Total profit (\$) columns of Table VII, respectively. Model 2 is solved by Price (\$) 225.58 240.26 229.65 -- th CPLEX with the objective function value given in the 4 Power output (MW) 154.49 40.0 40.0 -- Model 3 Profit (\$) 34386.4 9490.4 9066.0 52942.8 column of Table VII. Models 1-3 are MIBLP, MILP, and Power output (MW) 290.0 -60.0 101.71 -- MINLP models, respectively, as mentioned in Section III. Model 2 Profit (\$) 64548.2 -14595.6 23052.6 73005.2 These three models are solved using different numbers of time steps (i.e., ). The stopping criteria for both the BARON and CPLEX solvers are set to a maximum time of 8 hours and a gap of 0.1%; the solving process will terminate as soon as one of the two criteria is met. Table VII shows that the objective function value obtained from Model 2 is higher than that from Models 1 and 3 for all three different numbers of steps. Note that a larger objective https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 15 (a) (a) -100 0 5 10 15 20 25 Constant-temperature model Accurate value (b) (b) 0 5 10 15 20 25 70 Constant-temperature model (c) Accurate value (c) 0 5 10 15 20 25 Fig. 13. The electricity market price (a) and the solution obtained by Models 1- 3 in the 24-time-step case for the power output (b) and pressure (c). Hour (h) Fig. 14. The solution for the self-scheduling problem: (a) the power output 2) Comparison Between the Bilinear and Constant- obtained from the constant-temperature model, (b) the temperature obtained temperature Cavern Models from the constant-temperature model and the corresponding accurate value, and (c) the pressure obtained from the constant-temperature model and the To show the advantage of the proposed bilinear cavern model corresponding accurate value. over the constant-temperature model, both models are used to solve the SSP and comparative analysis is given as follows. The F. Cavern Efficiency results, including the power output, temperature, and pressure, The efficiency of the cavern is analyzed in this subsection. are given in Fig. 14. In Figs. 14b and 14c, the dots represent the The internal energy is calculated from , i.e., the product of temperature and pressure obtained from the SSP using the mass of air , a constant specific heat value , and temperature constant-temperature cavern model as detailed in Section III-C. . Let the cavern efficiency be the ratio of the internal energy Given the power output obtained from the SSP using the of all the air discharged out of the cavern in a period of time to constant-temperature cavern model as input, the SSP using the the internal energy of all the air charged into the cavern in the bilinear cavern model is used to calculate the temperature and same period of time. Note that the temperature of the air pressure as shown in the circles in Figs. 14b and 14c. That is, charged into the cavern is always equal to 50 °C because of the the circles given in Figs. 14b and 14c are the accurate values for use of aftercoolers. the charging/discharging process given in Fig. 14a. Fig. 14b The initial temperatures of cavern air in the first day are set shows that the temperature during the charging/discharging to 35, 40, and 45 °C, respectively. For each setting, Model 3 for process varies in a range of about 30 K, which is quite different the SSP of CAES in one day is solved; the internal energies of from constant. In Fig. 14c, the pressure obtained from the air injected into and released out of the cavern on this day are constant-temperature cavern model varies from 46 to 56.43 bar nd th given in the 2 and 4 columns of Table IX, respectively; and (shown in dots); however, the pressure actually varies from the heat transferred from the cavern wall to the cavern air on 40.68 to 63.44 bar (shown in circles). At both hours 21 and 22, rd this day is given in the 3 column of Table IX. Note that the the pressure obtained from the constant-temperature cavern rd positive (negative) values in the 3 column of Table IX indicate model is 46 bar; however, the pressure actually drops down to that less (more) heat is transferred to the cavern wall than is 40.68 and 40.79 bar, respectively, at these two time points. That received from the cavern wall. is, the pressure obtained from the constant-temperature cavern The cavern efficiency for each setting is given in the last model is quite inaccurate and the actual pressure drops far column of Table IX. Table IX shows that the cavern efficiency below the minimum bound of the normal operating range (i.e., is higher than 1 when the initial temperature is 35 °C. This is 46 bar as shown in the dashed line in Fig. 14c). This inaccurate because the temperature of the cavern wall (40 °C) is higher constant-temperature cavern model, however, has been adopted than 35 °C and thus heat is transferred to the cavern air from the in a variety of power system application areas, e.g., [31], [32], cavern wall. That is, the cavern air gains energy from the cavern [33], [34]. The proposed bilinear cavern model can replace the wall and therefore the cavern efficiency is higher than 1. But constant-temperature cavern model and is suitable for these when the initial temperature is 40 or 45 °C, the cavern power system applications and beyond, which shows the value efficiency is lower than 1; this is because the temperature of of the proposed bilinear cavern model. cavern air is equal to or higher than that of the cavern wall and therefore heat is transferred to rather than received from the https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 16 cavern wall. That is, the cavern air loses energy to the cavern compressed air energy storage. The accuracy of the proposed wall and therefore the cavern efficiency is lower than 1. bilinear model decreases as the time interval increases. For time To determine the average cavern efficiency for a longer intervals between 1 second and 10 minutes, equal to 20 minutes, period of time, the efficiencies for 10 consecutive days for each and 60 minutes or longer, the bilinear cavern model has high, nd th setting are plotted in Fig. 15. For each of the 2 to 10 days, moderate, and relatively low accuracy, respectively. Simulation the initial temperature of the cavern air in the next day is set to results also show that heat transfer has an obvious and be the temperature of the cavern air at the end of the previous measurable effect on the variation of temperature and pressure day and then the SSP of CAES in the day is solved to calculate of the air in the cavern. Therefore, it is necessary to consider the cavern efficiency for that day. The figure shows that the heat transfer in the cavern model. The constant-temperature cavern efficiencies on the first day are quite different, but are cavern model is also shown to be inaccurate. nd th close to one another for the 2 -10 days. The average The self-scheduling problems of compressed air energy efficiency for all days is 0.9572 and the average efficiency for storage using different cavern models are solved. The all days excluding the first day is 0.9579. Therefore, the average simulation results show that the self-scheduling problem of cavern efficiency is about 0.96. compressed air energy storage using the proposed bilinear cavern model can be straightforwardly converted into a mixed- integer linear programming model that is easier to solve than TABLE IX the mixed-integer nonlinear programming. The self-scheduling INTERNAL ENERGY OF AIR CHARGED IN AND DISCHARGED OUT, HEAT TRANSFERRED FROM CAVERN WALL, AND THE CAVERN EFFICIENCY problem of compressed air energy storage using the constant- Initial Internal Heat Internal Cavern temperature cavern model can result in infeasible solutions. temperature energy of transferred energy of efficiency However, by properly setting the time interval, the proposed of cavern air air in (J) from cavern air out (J) bilinear cavern model is accurate and suitable for use in power (°C) wall (J) system optimization problems and is superior to the existing 35 3.45E+15 0.14E+15 3.59E+15 1.04 40 3.40E+15 -0.17E+15 3.23E+15 0.95 highly nonlinear cavern model and the constant-temperature 45 3.42E+15 -0.48E+15 2.94E+15 0.86 cavern model. APPENDIX = ( − 2) , = , = = , = ℎ ̇ − ℎ ̇ , = , ( ) = − 1 , = , ( ) = − 1 , = , = , = , = Δ ( ) , Fig. 15. Cavern efficiencies for 10 consecutive days using different initial temperatures for the first day. = / , = ( ) / , V. CONCLUSION ( ) = − 2 Δ− ℎ Δ − ℎ − ℎ , This paper proposed an accurate bilinear cavern model for = Δ( − 1)( ) − (ℎ + ℎ )( − 2), compressed air energy storage based on the ideal gas law and ( ) ( )( ) c = Δ + ℎ Δ − Δ − 1 − the first law of thermodynamics. The accuracy of the proposed ( )( ) bilinear cavern model is verified via comparison with both an ℎ + ℎ 3 − , accurate analytical model in the literature and two sets of field- . . = ℎ Δ ̇ − ℎ Δ̇ − ℎ Δ − ℎ ̇ + measured data. ℎ ̇ , Simulation results show that the mean absolute percentage c = − ( − 2), error (mean absolute error) between the bilinear cavern model and the accurate analytical model for pressure (temperature) are . = Δℎ ̇ − Δℎ ̇ + ℎ Δ − 3 − no more than 0.82% and 0.36 bar (0.82% and 2.58 °C) for a variety of conditions. Note that the errors are usually less than ( ) ( )( ) the worst-case values given above. Simulation results also show = − 1 Δ− ℎ − 0.5 − 2 ℎ + ℎ − that the pressure (temperature) results obtained from the Δ ℎ / − ℎ , bilinear cavern model can well match the pressure (temperature) c = ( Δ( ) + ℎ ), from the two sets of field-measured data considered, i.e., ( ( )( ) ) = Δ 1 − + ℎ Δ+ Δℎ , datasets I and II, with a relative error of 0.47 and 1.01% (0.19 and 3.16%), respectively, in the normal operating mode of = ℎ ̇ − ℎ ̇ + ℎ , Cavern efficiency https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 17 . [10] M. Budt, D. Wolf, R. Span, J. Yan, “A review on compressed air energy = − ℎ ̇ + ℎ − ℎ ̇ Δ/ + storage: Basic principles, past milestones and recent developments”, . . Applied Energy, vol. 170, pp. 250-268, 2016. ℎ ̇ − ̇ + 0.5 ( − 2)ℎ ̇ − [11] J. Wang, K. Lu, L. Ma, J. Wang, M. Dooner, S. Miao, J. Li, D. Wang, "Overview of Compressed Air Energy Storage and Technology ̇ , Development”, Energies, vol. 10, no. 7, pp. 1-22, 2017. = Δℎ ̇ − ̇ , [12] X. Luo, J. Wang, C. Krupke, Y. Wang, Y. Sheng, J. Li, Y. Xu, D. Wang, S. Miao, H. Chen, “Modelling study, efficiency analysis and optimisation ( ( ) ( ) ) = ℎ + ℎ − − 1 Δ− ℎ , of large-scale adiabatic compressed air energy storage systems with low- = ℎ , temperature thermal storage”, Applied Energy, vol. 162, pp. 589–600, . . c = ℎ ̇ − ℎ ̇ − ℎ + ℎ ̇ − [13] E. Barbour, D. Mignard, Y. Ding, Y. Li, “Adiabatic compressed air energy storage with packed bed thermal energy storage”, Applied Energy, vol. 155, ℎ ̇ , pp. 804-815, 2015. [14] H. Peng, R. Li, X. Ling, H. Dong, “Modeling on heat storage performance = ℎ + ℎ ̇ − ℎ ̇ , of compressed air in a packed bed system”, Applied Energy, vol. 160, pp. 1–9, 2015. c = − Δ+ , [15] N. Hartmann, O. Vöhringer, C. Kruck, L. Eltrop, “Simulation and analysis of different adiabatic compressed air energy storage plant configurations”, = − ℎ , Applied Energy, vol. 93, pp. 541-548, 2012. ( ( ) ) = Δℎ + ℎ + Δℎ + ℎ , [16] Y. Zhang, Y. Xu, X. Zhou, H. Guo, X. Zhang, H. Chen, “Compressed air = − Δℎ , energy storage system with variable configuration for accommodating large-amplitude wind power fluctuation”, Applied Energy, vol. 239, pp. = ℎ , 957-968, 2019. = − Δℎ + ℎ ̇ − ℎ ̇ , [17] J.D. Wojcik, J. Wang, “Feasibility study of Combined Cycle Gas Turbine (CCGT) power plant integration with Adiabatic Compressed Air Energy = ℎ + ℎ ̇ − ℎ ̇ , Storage (ACAES)”, Applied Energy, vol. 221, pp. 477-489, 2018. . [18] I. Ortega-Fernández, S. A. Zavattoni, J. Rodríguez-Aseguinolaza, B. = − ̇ − ̇ , D'Aguanno, M.C. Barbatob, “Analysis of an integrated packed bed thermal energy storage system for heat recovery in compressed air energy ( ) = Δℎ + ℎ ̇ − ̇ , storage technology”, Applied Energy, vol. 205, pp. 280-293, 2017. ( ) [19] X. Zhang, Y. Xu, X. Zhou, Y. Zhang, W. Li, Z. Zuo, H. Guo, Y. Huang, = , c = − , H. Chen, “A near-isothermal expander for isothermal compressed air energy storage system”, Applied Energy, vol. 225, pp. 955-964, 2018. ( ) c = − , = − − − 1 , [20] B. Bollinger, “Technology Performance Report - SustainX Smart Grid Program-Demonstration of Isothermal Compressed Air Energy Storage to ( ) = − , = 1 − + . Support Renewable Energy Production”, Apr. 1, 2015. Available:https://www.smartgrid.gov/project/sustainx_inc_isothermal_co mpressed_air_energy_storage.html. [21] A. Arabkoohsar, M. Dremark-Larsen, R. Lorentzen, G.B. Andresen, ACKNOWLEDGEMENTS “Subcooled compressed air energy storage system for coproduction of heat, The work was supported in part by the Natural Sciences and cooling and electricity”, Applied Energy, vol. 205, pp. 602-614, 2017. Engineering Research Council (NSERC) of Canada and the [22] B. Kantharaj, S. Garvey, A. Pimm “Compressed air energy storage with Saskatchewan Power Corporation (SaskPower). liquid air capacity extension”, Applied Energy, vol. 157, pp. 152–164, [23] S. Succar, D. C. Denkenberger, R.H. Williams, “Optimization of specific REFERENCES rating for wind turbine arrays coupled to compressed air energy storage”, Applied Energy, vol. 96, pp. 222-234, 2012. [1] Global wind energy council, Global Wind Energy Outlook 2016, [24] X. Luo, M. Dooner, W. He, J. Wang, Y. Li, D. Li, O. Kiselychnyk, Available: http://www.gwec.net. “Feasibility study of a simulation software tool development for dynamic [2] U. S. Department of Energy, “EPRI-DOE Handbook of Energy Storage modelling and transient control of adiabatic compressed air energy storage for Transmission & Distribution Applications: Final Report,” 2003. with its electrical power system applications”, Applied Energy, vol. 228, [3] X. Luo, J. Wang, M. Dooner, and J. Clarke, “Overview of current pp. 1198-1219, 2018. development in electrical energy storage technologies and the application [25] S. Briola, P. D. Marco, R. Gabbrielli, J. Riccardi, “A novel mathematical potential in power system operation”, Applied Energy, vol. 137, pp. 511- model for the performance assessment of diabatic compressed air energy 536, 2015. storage systems including the turbomachinery characteristic curves”, [4] International Energy Agency, “Technology Roadmap Energy Storage,” Applied Energy, vol. 178, pp. 758-772, 2016. [Online]. Available: https://www.iea.org, accessed on 19 March 2014. [26] T. Das and J.D. McCalley, “Compressed air energy storage”, Educational [5] World Energy Council, World Energy Resources – Hydropower 2016, Chapter, Iowa State Univ. 2012. [Online] https://www.worldenergy.org [27] R. Kushnir, A. Ullmann, A. Dayan, “Thermodynamic models for the [6] BBC Brown Boveri, Huntorf Air Storage Gas Turbine Power Plant, temperature and pressure variations within adiabatic caverns of Available: http://www.solarplan.org/Research/BBC_Huntorf_engl.pdfS. compressed air energy storage plants”, J. Energy Resources Technology, Succar and R. H. Williams. Compressed Air Energy Storage: Theory, vol. 134, no. 2, pp. 1-10, 2012. Resources, and Applications for Wind Power. Princeton, NJ: Princeton [28] M. Raju, and S.K. Khaitan, “Modeling and simulation of compressed air University. 2008. storage in caverns: A case study of the Huntorf plant”, Applied Energy, [7] U. S. Department of Energy / National Energy Technology Laboratory vol. 89, no. 1, pp. 474-481, 2012. (NETL), “Technical and Economic Analysis of Various Power Generation [29] C. Xia, Y. Zhou, S. Zhou, P. Zhang, F. Wang, “A simplified and unified Resources Coupled with CAES Systems: Final Report”, May 2011. analytical solution for temperature and pressure variations in compressed [8] Power South Energy Cooperative, Compressed air energy storage-- air energy storage caverns”, Renewable Energy, vol. 74, pp. 718-726, 2015. McIntosh Power Plant, Alabama, U. S. [Online]. http://www.powersouth. [30] W. He, X. Luo, D. Evans, J. Busby, S. Garvey, D. Parkes, J. Wanga, com/files/CAES%20Brochure%20[FINAL].pdf “Exergy storage of compressed air in cavern and cavern volume estimation [9] Rocky Mountain Power, “Alberta Saskatchewan Intertie Storage (ASISt)”, of the large-scale compressed air energy storage system”, Applied Energy, [Online] http://rockymountainpower.ca/ASISt.html vol. 208, pp. 745-757, 2017. [31] H. Khani, M.R.D. Zadeh, A.H. Hajimiragha, “Transmission congestion relief using privately owned large-scale energy storage systems in a https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 18 competitive electricity market”, IEEE Trans. Power Syst., vol. 31, no. 2, pp. 1449-1458, 2016. [32] S. Shafiee, H. Zareipour, A.M. Knight, N. Amjady, B. Mohammadi- Ivatloo, “Risk-constrained bidding and offering strategy for a merchant compressed air energy storage plant”, IEEE Trans. Power Syst., vol. 32, no. 2, pp. 946-957, 2017. [33] H. Daneshi and A.K. Srivastava, “Security-constrained unit commitment with wind generation and compressed air energy storage”, IET Gener. Transm. Distrib., vol. 6, no. 2, pp. 167-175, 2012. [34] X. Luo, J. Wang, M. Dooner, J. Clarke, C. Krupke, “Overview of current development in compressed air energy storage technology”, Energy Procedia, vol. 62, pp. 603-611, 2014. [35] T.D. Eastop and A.C. McConkey. Applied Thermodynamics for Engineering Technologists. Delhi, India: Pearson Education, 2009. [36] Gilbert Strang, Calculus, Wellesley, MA, USA: Wellesley-Cambridge Press, 2010. [37] I. Grossmann, “Review of nonlinear mixed–integer and disjunctive programming techniques”, Optimization and Engineering, vol. 3, pp. 227- 252, 2002. [38] S. Sager, C.M. Barth, H. Diedam, M. Engelhart, J. Funke, “Optimization as an analysis tool for human complex problem solving”, SIAM J. Optim, vol. 21, no. 3, pp. 936-959, 2011. [39] P.M. Castro, “Tightening piecewise McCormick relaxations for bilinear problems”, Computers and Chemical Engineering, vol. 72, pp. 300-311, [40] A. Zare, C.Y. Chung, J.P. Zhan, S.O. Faried, “A distributionally robust chance-constrained MILP model for multistage distribution system planning with uncertain renewables and loads”, IEEE Trans. Power Systems, vol. 33, no. 5, pp. 5248-5262, 2018. [41] E. Johnson, G.L. Nemhauser, M. Savelsbergh, “Progress in linear programming based algorithms for integer programming: An exposition”, INFORMS J. Computing, vol. 12, no. 1, pp. 2-23, 2000. [42] IBM, IBM ILOG CPLEX V12.1: User's manual for CPLEX, http://www.ibm.com, 2017. [43] Gurobi Optimization, Inc., Gurobi Optimizer Reference Manual, http://www.gurobi.com, 2017. [44] H.Y. Yamin and S.M. Shahidehpour, “Self-scheduling and energy bidding in competitive electricity markets”, Electric Power Syst. Res., vol. 71, no. 3, pp. 203-209, 2004. [45] G. Pedrick, “Compressed air energy storage engineering and economic study - final report by New York state energy research and development authority”, 2009, [Online]. https://www.nyserda.ny.gov [46] N.V. Sahinidis, BARON 17.8.9: Global Optimization of Mixed-Integer Nonlinear Programs, User's manual, 2017. [47] Y. Zimmels, F. Kirzhner, B. Krasovitski, “Design criteria for compressed air storage in hard rock”, Energy & Environment, vol. 13, no. 6, pp. 851- 872, 2002. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Electrical Engineering and Systems Science arXiv (Cornell University)

# An Accurate Bilinear Cavern Model for Compressed Air Energy Storage

, Volume 2019 (1903) – Mar 16, 2019
18 pages

Loading next page...

/lp/arxiv-cornell-university/an-accurate-bilinear-cavern-model-for-compressed-air-energy-storage-07Pgm45GHb

# References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

ISSN
0306-2619
eISSN
ARCH-3348
DOI
10.1016/j.apenergy.2019.03.104
Publisher site
See Article on Publisher Site

### Abstract

https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 1 An Accurate Bilinear Cavern Model for Compressed Air Energy Storage a,b a a a Junpeng Zhan , Osama Aslam Ansari , Weijia Liu , C. Y. Chung Department of Electrical and Computer Engineering, University of Saskatchewan, Saskatoon, SK S7N 5A9, Canada Sustainable Energy Technologies Department, Brookhaven National Laboratory, Upton, NY 11973, U.S.A. Mass of air in virtual container 2 as shown in Fig. 3 Abstract—Compressed air energy storage is suitable for large- (kg) scale electrical energy storage, which is important for integrating renewable energy sources into electric power systems. A typical Mass of air charged into the cavern (kg) compressed air energy storage plant consists of compressors, Number of time steps expanders, caverns, and a motor/generator set. Current cavern Pressure of the air charged into the cavern (bar) models used for compressed air energy storage are either accurate Pressures in virtual states as shown in Fig. 2, = 1,2 but highly nonlinear or linear but inaccurate. The application of (bar) highly nonlinear cavern models in power system optimization Maximum and minimum pressures in a cavern for problems renders them computationally challenging to solve. In optimal operation of compressed air energy storage this regard, an accurate bilinear cavern model for compressed air (CAES), respectively (bar) energy storage is proposed in this paper. The charging and discharging processes in a cavern are divided into several Pressure of the air in the cavern after the charging, real/virtual states. The first law of thermodynamics and ideal gas discharging, and idle processes for =2, 3, and 4, law are then utilized to derive a cavern model, i.e., a model for the respectively (bar) variation of temperature and pressure in these processes. Time step Thereafter, the heat transfer between the air in the cavern and the Surface area of the cavern wall (m ) cavern wall is considered and integrated into the cavern model. By , Operational costs of charging and discharging power subsequently eliminating several negligible terms, the cavern in CAES, respectively (\$/MWh) model reduces to a bilinear model. The accuracy of the bilinear Maximum and minimum charging power of CAES, cavern model is verified via comparison with both an accurate respectively (MW) nonlinear model and two sets of field-measured data. The bilinear cavern model can be easily linearized and is then suitable for Maximum and minimum discharging power of integration into optimization problems considering compressed CAES, respectively (MW) air energy storage. This is verified via comparatively solving a self- Total internal energy (J) scheduling problem of compressed air energy storage using Specific air constant (J/(kg K)) different cavern models. Temperature of the air in the cavern after the Index Terms—Bilinear cavern model; compressed air energy charging, discharging, and idle processes for =2, 3, storage (CAES); heat transfer; ideal gas law; thermodynamics. and 4, respectively (K) Temperature of the cavern wall (K) Temperature of the air charged into the cavern (K) NOMENCLATURE Temperatures in virtual states as shown in Fig. 2, = Parameters 1,2 (K) a,ht Both adiabatic process and heat transfer are , Volumes of virtual containers as shown in Figs. 2 and considered (superscript) 3 (m ) Parameters, = 1,2,3, ⋯ ,13, given in the Appendix Volume of a cavern (m ) / Coefficients used to model the relationship between Work (J) charging/discharging power and mass flow rate in/out Average air density in a cavern (kg/m ) Parameters, = 0,1, representing the left-hand side Electricity price (\$/MWh) of (3) and (10), respectively, given in the Appendix Δ Time interval (s) Parameters, = 2,3, ⋯ ,32, given in the Appendix Δ Change in internal energy (J) Constant volume specific heat (J/(kg K)) 0,1,2, ⋯ , ( − 1) ht Heat transfer (superscript) Ω 1,2,3, ⋯ , Heat transfer coefficient (W/(m K)) A constant equal to 1.4 Variables Parameters used in (35), (36), (47), and (48), = 1,2,3,4 ̇ Mass flow rate charged into a cavern (kg/s) ̇ Mass flow rate discharged out of a cavern (kg/s) Mass of air in the cavern (kg) E-mail addresses: zhanjunpeng@gmail.com, oa.ansari@usask.ca, liuweijiamarcel@gmail.com, c.y.chung@usask.ca). https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 2 Pressure of air in the cavern (bar) about 75% of the continental U.S. has geologic sites suitable for ( ) CAES [7], [8]. Northern Europe is also replete with suitable salt Pressure after a (xx) process, where (xx) can be ‘ch’, ,() ‘dch’, and ‘idl’, which represent charging, deposits. For example, nearly 500 salt caverns are currently being used for natural gas storage. That is, it is feasible to install discharging, and idle, respectively (bar) Temperature of air in the cavern (K) CAES in many different locations, which makes it a promising ( ) large-scale energy storage technology. Temperature after a (xx) process, where (xx) can be ,() ‘ch’, ‘dch’, and ‘idl’, which represent charging, Discharg ing circuit discharging, and idle, respectively (K) Charging circuit , Charging and discharging power, respectively (MW) Binary variable indicating the charging and M/G 3 4 6 5 discharging processes, respectively 2 1 2 I. INTRODUCTION Air Fuel 8 Fuel Air Energy storage technologies are critical to electric power Air 1: Motor or Generator 2: Clutches systems, especially considering that the penetration of 3: High pressure turbine Charging Discharging renewable generation is growing rapidly, e.g., the share of wind 4: Low pressure turbine process process 5: Low pressure compressor power in global electricity generation will increase from 4% in 6: High pressure compresso r 2015 to 25-28% in 2050 [1]. Energy storage can provide various 7: Cavern (also represents Repository containers 3 or 5 in Fig. 2 kinds of services [2], [3], e.g., electric energy time-shift, electric (underground cavern) and container 3 in Fig. 3) supply capacity, regulation, power reliability, etc. The current 8: Represent virtual container 1 in Fig. 2 global installed energy storage capacity is approximately 141 GW, and an estimated 310 GW of additional capacity is needed Fig. 1. Schematic of the Huntorf compressed air energy storage plant. in the United States, Europe, China, and India [4] to support the massive increase of renewable generation in the future. Two Two commercialized CAES plants are currently in operation. The world’s first CAES plant was installed in Huntorf, types of large-scale energy storage currently exist, i.e., pumped- hydro power storage and compressed air energy storage Germany in 1978 [6]. The power ratings of its charging and discharging processes are 60 and 290 MW, respectively. The (CAES), which can be cost-effectively installed at the electricity grid scale. Hydropower technologies are mature, and second commercialized CAES plant is the McIntosh plant in McIntosh, Alabama, U.S. [8], which became operational in many sites that were feasible for constructing pumped-storage hydropower already have hydropower plants in place [5]. 1991. The power rating of its charging process is 50 MW and it can produce an output of 110 MW electricity for up to 26 hours. Therefore, this paper focuses on CAES technology. A CAES plant consists of compressors, turbines, a Due to several benefits offered by CAES, a number of CAES plants are being constructed for efficient utilization of motor/generator set, and large repositories, e.g., underground salt caverns. Fig. 1 shows a schematic of the Huntorf CAES renewable energy sources. For example, a feasibility study for a 160 MW CAES plant near the Saskatchewan-Alberta border plant to depict the main components of CAES [6]. CAES operates in one of three modes, i.e., charging, discharging, and in Canada [9] was finished in late 2018. This proposed CAES plant is expected to be combined with the interconnection idle. When charging, the motor/generator set acts as a motor that is connected to the compressors via a clutch, as shown in between the Saskatchewan and Alberta power grids. From 2009 to 2013, Pacific Gas & Electric received US\$50 million in the ‘Charging circuit’ part of Fig. 1. Low-cost electricity (e.g., off-peak electricity or wind power) is usually used by the motor funding for a demonstration project to validate the design, performance, and reliability of a 300 MW CAES plant in Kern to compress air to high pressure for storage in a large repository. When discharging, the motor/generator set acts as a generator County, California [10]. Several further examples are provided in [10], [11]. that is connected to the turbines via another clutch, as shown in the ‘Discharging circuit’ part of Fig. 1. The compressed air is The two existing CAES plants in Huntorf and Alabama are diabatic CAES (D-CAES), and also referred to as the first released from the repository and then combusted with fuel (e.g., natural gas) to drive the turbines. When idling, no air is injected generation CAES. Second generation CAES include adiabatic CAES (A-CAES) [12] [13], [14], [15] [16], [17], [18], into or released from the repository. Multiple stages of compressors (turbines) are typically used instead of a single isothermal CAES (I-CAES) [19], [20], and hybrid CAES [21], [22]. In the D-CAES, the heat generated in the compression stage to increase the efficiency, e.g., low- and high-pressure compressors (turbines) are used in the Huntorf CAES plant as process is wasted and an external heat source is required in the expansion process. In A-CAES, the heat generated in the shown in Fig. 1. CAES is a high power and high energy storage technology compression process is captured, stored, and then used in the expansion process. I-CAES prevents the temperature from and has relatively low capital, operational, and maintenance costs [3]. The power rating of a large-scale CAES plant can changing in the compression and expansion devices. Pacific Northwest National Laboratory (PNNL) and EPRI consider A- reach 300 or even 1000 MW and the rated energy capacity can reach 1000 or even 2860 MWh [3]. Different definitions of CAES and hybrid CAES to be the most suitable and promising CAES technologies [10]. For a complete review of CAES efficiencies for CAES are discussed in [7], e.g., the round-trip efficiency of the CAES typically ranges from 66 to 82%. technologies, readers are referred to [10], which provides a comprehensive classification and comparison of different According to the Electric Power Research Institute (EPRI), https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 3 CAES technologies as well as several challenging issues value terms and using several linearization methods such as relevant for CAES research and development. Newton’s generalized binomial theorem and the first-order Air reservoir is a fundamental part of the CAES as it Taylor series approximation [36]. This original bilinear model significantly affects the power and energy rating as well as the is derived step-by-step in this paper. The advantages of the total construction and operation costs [11]. The reservoirs can bilinear model over the existing model types mentioned above be above-ground containers or underground caverns, including are twofold: 1) it is accurate, as will be verified in this paper, salt, hard rock, and porous rock layer caverns [11], [23]. The and 2) it is suitable for integration in power system optimization existing commercial CAES plants mentioned above both use problems as explained in the following two paragraphs. underground salt caverns for air storage. The proposed cavern To consider CAES in an optimization problem, binary model is general and can be applied to the different kinds of variables are required to indicate the charging, discharging, or reservoirs mentioned above. idle status. That is, an optimization problem considering CAES Some research has been conducted to model compressors and is a mixed-integer programming problem. If the nonlinear turbines [24], [25], [26]. However, the CAES cavern models cavern models given in [27], [28], [29] are used for CAES, an available in the literature [11], [27], [28], [29], [30] are either optimization problem considering CAES becomes a mixed- too complicated or inaccurate and, as such, are not suitable for integer nonlinear programming (MINLP) problem. MINLP real applications. Therefore, this paper focuses on the cavern problems are usually difficult to be accurately solved in an modeling of CAES, i.e., modeling the temperature and pressure acceptable computational time [37], [38], and therefore variations of cavern air. Thermodynamic properties, such as nonlinear cavern models such as those given in [27], [28], [29] variations of temperature and pressure in the caverns and the are not suitable to be employed in an optimization problem. If heat transfer between caverns and their surroundings, are the proposed bilinear cavern model is used for CAES, an important factors that affect the overall plant operation and optimization problem considering CAES becomes a mixed- performance [27], [28], [29]. Two kinds of cavern models for integer bilinear programming (MIBLP) problem, which can CAES are currently described in the literature. easily be converted into a mixed-integer linear programming The first kind is models that are accurate but highly nonlinear (MILP) problem [39], [40]. MILP problems are relatively easier [27], [28], [29]. In [27], complex and simplified real gas models to solve than MINLP problems [37], [41]. MILP problems can are developed for an adiabatic cavern for CAES, both of which also be efficiently solved by several mature, off-the-shelf adequately represent the thermodynamic properties of the air. commercial solvers, e.g., CPLEX [42] and Gurobi [43]. Reference [28] developed an accurate dynamic simulation Therefore, the bilinear cavern model is more suitable for model for a CAES cavern that incorporates an accurate heat integration into an optimization problem compared to the transfer model; the heat transfer is shown to play an important highly nonlinear cavern models given in [27], [28], [29]. The role in the thermodynamic behavior of the cavern and therefore above discussion serves as the main motivation to propose the the model in [28] can accurately simulate the actual cavern bilinear model. behavior. In [29], a simplified and unified analytical solution A typical example of the previously mentioned optimization considering heat transfer is proposed for temperature and problem is as follows. A CAES plant can act as an independent pressure variations in CAES caverns. This model is validated participator in electricity markets. In this case, the self- using data from the Huntorf plant trial tests and the results scheduling of a CAES plant is an important optimization calculated from the models in [27] and [28], demonstrating that problem that needs to be solved to maximize its arbitrage the solution in [29] is capable of adequately calculating the revenue obtained from selling electricity to and buying thermodynamic behavior of CAES caverns. Note that the electricity from the market [34], [44], [45]. In this paper, the models described in [27], [28], and [29] are accurate but highly day-ahead self-scheduling problem (SSP) of CAES is used to nonlinear. demonstrate the advantage of using the proposed bilinear model The second kind of cavern model assumes that the air in an optimization problem compared to the highly nonlinear temperature in the cavern is constant. This kind of model has cavern models given in [29]. For the sake of comparison, both been adopted in different power system operation problems, the proposed bilinear cavern model and the nonlinear cavern e.g., transmission congestion relief [31], bidding and offering model in [29] are integrated into the SSP of CAES; both SSPs strategy [32], and unit commitment [33]. An overview of are MINLP problems and can be solved by an MINLP solver, different applications of technologies relevant to CAES is e.g., BARON [46] (note that MIBLP is a kind of MINLP). available in [34]. The constant-temperature model is linear but Furthermore, the SSP of CAES using the proposed bilinear inaccurate, which can result in non-optimal or even infeasible model is an MIBLP model; the MIBLP model is converted into solutions in practice. an MILP model, which is then solved by an MILP solver, e.g., Therefore, all existing CAES applications use simple but CPLEX. inaccurate cavern models. To achieve a balance between In summary, the contributions of this paper include 1) accuracy and complexity, this paper proposes an original and proposing a novel accurate bilinear cavern model of CAES that novel bilinear cavern model. The charging (discharging) is superior to existing linear but inaccurate or accurate but process is divided into four (two) real/virtual states. The ideal highly nonlinear cavern models from an optimization point of gas law and the first law of thermodynamics [35] are used to view, 2) verifying the accuracy of the proposed bilinear cavern model the pressure/temperature relationships between different model via comparison with both an accurate model and field- charging/discharging states. The heat transfer between the measured data, and 3) using the SSP of CAES to verify the cavern air and the cavern wall is also considered. A bilinear superiority of using the bilinear cavern model compared to a cavern model is then obtained by ignoring some very-small- highly nonlinear cavern model in the optimization problem. https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 4 The reminder of this paper is organized as follows. Section Then, let the air in both containers 2 and 3 go into container II details the deduction of the accurate bilinear cavern model. 4. The volume of (mass of air in) container 4 is set to the sum Section III presents the SSP of CAES and flowcharts for using of the volumes of (masses of air in) containers 2 and 3, i.e., the bilinear cavern model. Section IV verifies the effectiveness container 4 can be seen as a combination of containers 2 and 3. of the bilinear cavern model. Section V presents the conclusions Note that the purpose of using virtual containers 2 and 4 is to drawn from the results. let the work be zero during the process of merging the air in containers 2 and 3 into container 4. Last, let the air in container 4 go into container 5, which is an II. ACCURATE BILINEAR CAVERN MODEL FOR COMPRESSED adiabatic compression process as the mass of air does not AIR ENERGY STORAGE change and the volume decreases. The rest of this subsection The cavern of a CAES plant can operate in either constant- details the deduction of the model for the charging process. volume or constant-pressure mode. In this paper, constant- 1) State 1 State 2 volume caverns are considered as they are used in existing The transfer of air from containers 1 to 2 is an adiabatic CAES plants [10]. process and the mass of air does not change. According to the ideal gas law [35] for the air in container 2, one can obtain A. Charging Process = ̇ Δ . (2) In the charging process, a certain amount of air is injected by compressors into a cavern, as shown in Fig. 1. To facilitate the According to the temperature-pressure relationship for an model deduction given in Sections II-A1 to II-A3, the charging adiabatic process, the term / is constant from containers process is divided into four states and the air is assumed to be 1 to 2 [47] and therefore one can obtain stored in the (virtual) containers as shown in Fig. 2, where the five containers are indexed by the numbers in heptagons. The ( ) ( ) characteristics of the air in each container, including the = . (3) pressure, volume, temperature, and mass, are shown in each container in Fig. 2. The values of the underlined notations, i.e., Let represent the left-hand side of (3), i.e., = / . the pressure and temperature in containers 2, 4, and 5, are Note that the constant is known. can be determined unknown while the values of the other notations are known from (1), i.e., = ̇ Δ/ . This leaves only two from the actual CAES operation. unknown variables in (2) and (3), i.e., and . Therefore, Containers 1, 2, and 4 are virtual while containers 3 and 5 and can be obtained from (2) and (3): represent the cavern before and after the air from the compressors is injected into it, respectively. = ( ) / (4) = ( / ) . (5) 2) State 2 State 3 Now we consider the process of the air in both containers 2 and 3 going into container 4. The volume of container 4 is equal Fig. 2. Four states and five containers used for model deduction in the charging to the sum of the volumes of containers 2 and 3. In this process, process. the work is zero and the total internal energy [35] does not change. The change of the internal energy in containers 2 and 3 The air coming out of the compressors is assumed to be is ( − ) and ( − ) , respectively. stored in a virtual container, i.e., cylinder 8 in Fig. 1 and According to the first law of thermodynamics, i.e., = Δ + container 1 in Fig. 2. The conditions of container 1 represent , one can obtain the thermodynamic properties of air at the outlet of the ( ) ( ) − + − = 0. (6) compressors. First, let the air in virtual container 1 go into virtual container 2. The volume of container 2 is set such that Note that can be obtained from (5). is then the only the ratio of the volume of containers 2 to 3 is equal to the ratio unknown variable in (6) and can be expressed as of the mass of air in containers 2 to 3, i.e., ( ) ( ) = + / + . (7) / = ̇ Δ/ (1) According to the ideal gas law for the air in container 4 in where and represent the volumes of containers 2 and 3, Fig. 2, one can obtain respectively; represents the mass of air in container 3; and ( + )= ( + ) . (8) ̇ represents the flow rate of mass charged into the cavern, which is assumed to be constant during a period of time, Δ. By substituting (7) into (8), one can then obtain The mass of air coming out of the compressor during that period ( ) ( ) of time, denoted as , can be expressed as = ̇ Δ. The = = . (9) mass of air injected into the cavern is assumed to be equal to , i.e., there is no air leakage. is also the mass of air in Now and in container 4 have been obtained. both containers 1 and 2. https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 5 3) State 3 State 4 known initial status. When used in a multi-time-step From containers 4 to 5, the mass of air does not change and optimization problem, (17) and (18) are bilinear equations as the volume reduces from ( + ) to . This is an adiabatic and become decision variables. compression process. Thus, / is constant from containers B. Discharging Process 4 to 5 [47], i.e., In the discharging process, the air is released from the cavern, as shown in Fig. 1. To facilitate the model deduction given in ( ) ( ) = (10) the rest of this subsection, the cavern before discharging is divided into two containers, i.e., containers 1 and 2, while the According to the ideal gas law for the air in container 5 in cavern after discharging (cylinder 7 in Fig. 1) is represented as Fig. 2, one has container 3 in Fig. 3. The three containers in Fig. 3 are indexed by the numbers in heptagons. The characteristics of the air in = (̇ Δ + ) (11) each container, including the pressure, volume, temperature, Let represent the left-hand side of (10), i.e., = and mass, are shown in each container in Fig. 3. The values of the underlined notations, i.e., the pressure and temperature in ( ) / . Only two variables are unknown in (10) and container 3, are not known while the values of the other (11), i.e., and . Therefore, and can be obtained notations are known. from (10) and (11): The discharging process can be divided into two virtual steps. ( ) ( ) = ̇ Δ + / (12) First, the air in container 2 is taken out of the cavern. Then, the air in container 1 expands to the whole cavern, i.e., air goes ( ) = . (13) from containers 1 to 3. The deduction of the model for the second step is given in the rest of this subsection. Now and for container 5 have been obtained. By substituting ( , (5), (7), and (9) are needed to calculate ) into (12) and (13), these two equations can be rewritten as: = 1 + + (̇ Δ + ) ̇ Δ (14) Fig. 3. Three containers used for model deduction in the discharging process. ( ) = 1 + + ̇ Δ + ̇ Δ (15) The volume of container 2 is set such that the ratio of the where = and = . volume of container 2 to that of container 1 is equal to the ratio Equations (14) and (15) show that and are nonlinear of the mass of air in container 2 to that of container 1, i.e., functions of ̇ , which are linearized as follows. According to /( − )= /( − ). (19) Newton’s generalized binomial theorem [36], one has Note that the purpose of using virtual container 2 is to let the ( ) (1 + ) = ∑ = 1 + + + ⋯ (16) temperature and pressure of the air in container 1 be the same as the air in the cavern before discharging and to let air go from ( ) ⋯ ( ) containers 1 to 3. This means that step 2 is an adiabatic where = , can be any real number, and is expansion process. an integer. That is, 1 + in (14) can be expressed as Let ̇ represent the flow rate of mass discharged from the cavern, which is assumed to be constant during a period of time, ̇ ( )( ) ̇ 1 + ( − 1) + + ⋯ . Considering Δ. Then, the mass of air discharged from the cavern, denoted as , during that period of time can be expressed as = that ̇ Δ is much smaller than , the second and higher ̇ Δ. orders terms of (̇ Δ/ ) can be ignored. Then, (14) can be The air expansion from containers 1 to 3 is an adiabatic reformed as process. Then, / is constant from containers 1 to 3: ( ) = 1 + ( − 1) + ̇ Δ (17) ( ) = ( ) . (20) Note that the second term in (14) is replaced by ( ) ̇ Δ, which has negligible error as is much According to the ideal gas law for the air in container 3 in smaller than (e.g., =46~66× 10 Pascals and =1.04× Fig. 3, one has 10 for the Huntorf CAES plant) and ̇ Δ is much smaller = ( − ) . (21) than . Similarly, (15) can be reformed as There are only two variables unknown in (20) and (21), i.e., ( ) = 1 + ( − 2) + ̇ Δ (18) and . Therefore, and can be obtained from (20) and (21): When (17) and (18) are used in a single-time-step = (1 − ̇ Δ/ ) (22) optimization problem, is a linear function of ̇ in (17) and is a linear function of ̇ in (18) as and have a = (1 − ̇ Δ/ ) . (23) https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 6 Similar to the charging process, the above nonlinear where superscript ‘a,ht’ indicates that both the adiabatic process equations are linearized. According to Newton’s generalized and heat transfer are considered. binomial theorem, when ̇ Δ is much smaller than , (22) According to the first-order Taylor series approximation [36], ( ) ( ) ( ) and (23) can be respectively reformulated as i.e., = + × ( − ) , one can linearize ( ) ( ) and at as = (1 − ̇ Δ/ ) (24) ( ) ( ) ( )( ) = + − 1 ( − ) (33) = (1 − ( − 1)̇ Δ/ ) . (25) ( ) ( ) ( )( ) = + − 2 ( − ) (34) Similar to (17) and (18), (24) and (25) are linear (bilinear) equations when used in a one-step (multi-step) optimization where is a fixed value, i.e., = . The ̇ problem. and ̇ in (32) can be linearized using . . C. Charging Process Considering Heat Transfer ( ) ̇ = ̇ + ̇ − ̇ (35) In Sections II-A and II-B, the heat transfer between the . . ̇ = ̇ + (̇ − ̇ ) (36) cavern air and the cavern wall is not considered. However, the . . . . ̇ ̇ ̇ ̇ heat transfer plays an important role in the variation of the air where = , = , and the setting ̇ ̇ ̇ ̇ temperature/pressure in the cavern [28]. Therefore, the heat of ̇ and ̇ will be given in the beginning of Section IV. transfer is considered in this subsection and the following two By using (33)-(36), equation (32) can then be converted to be a subsections. In this subsection, the temperature as a function of bilinear form, i.e., (37). time is first deduced. The pressure as a function of time is then obtained via the ideal gas law. Last, the temperature/pressure as ( ) = + ̇ + ̇ + c ̇ + + a function of time is linearized to obtain a bilinear model. c + (37) According to [28], [29], the air density ( ) in the cavern and the cavern wall temperature ( ) can be assumed to be where - are constant coefficients given in the Appendix. constant and the heat transfer between the cavern air and the ( ) Equation (37) involves four variables, i.e., , ̇ , , cavern wall can be modelled as and and the first four terms are bilinear terms. Equation (37) represents the change of the temperature = ( − ) (26) during the charging process as a function of time and the charging mass flow rate ̇ , where both the adiabatic process where and the heat transfer process are considered. ℎ = ℎ + ℎ |̇ − ̇ | (27) According to the ideal gas law, one can obtain , , According to [28], ℎ and ℎ can be set to 0.2356 and 0.0149, ( ) ( ) = + ̇ ()/ (38) respectively. Equation (26) can be reformed as which can be expanded to (39) by substituting (32) therein. ()= ( − ) (28) ∫ ( ) ̇ ̇ ( ) ( ) = 1 + − 2 ( ̇ ) Equation (18) can be written as ( ) + ̇ ( ̇ ) ̇ ()= 1 + ( − 2) + ( ) ̇ . (29) + ℎ + ℎ ̇ − − ( − 2) ( ) By substituting (29) into (28), i.e., replacing on the right- ( ) − ℎ + ℎ ̇ ̇ (39) hand side of (28) by the right-hand side of (29), one can obtain The four terms in (39) are each on a separate line. The second ( ) ( ) = ∫ − 1 + − 2 − term of (39) can be replaced by ( ) ̇ / because / is small and ̇ is much smaller than . ( ) ̇ (30) The last term of (39) can be ignored because the values of both ℎ + ℎ ̇ / and / are small. where superscript ‘ht’ represents ‘heat transfer’. By solving the Again, according to the first-order Taylor series integral equation (30), one can obtain approximation [36], i.e., ()= ( )+ ( )× ( − ), ( ) ( ) = − + − 2 − one can linearize at : ( ) ( ) ̇ /2. (31) = + − (40) Now, by using (40), (39) can be reformed into a bilinear Adding (29) and (31) together gives form: ()= 1 + ( − 2) + ( ) ̇ + ()= + ̇ + c ̇ + ̇ + ℎ + ℎ ̇ − + ( − 2) − + + (41) ( ) ̇ (32) https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 7 ( ̇ ) ( ̇ ) ̇ where - are constant coefficients as given in the Appendix. ()= − ( − 1) + ( ) Equation (41) involves four variables, i.e., , ̇ , , ( ) ( ) ̇ ̇ ( ) ( − + − and , and the first four terms are bilinear terms. 1) (51) D. Discharging Process Considering Heat Transfer In this subsection, the temperature as a function of time is ̇ ( ) Note that − ̇ in the third term of (51) can be first deduced. The pressure as a function of time is then obtained via the ideal gas law. Last, the temperature/pressure as a replaced by because ̇ is much smaller than . function of time is linearized to obtain a bilinear model. Equation (51) can then be reformed as Equation (25) can be written as ()= ( − ̇ ) ()= − ( − 1) . (42) + (( − )( − ) + 0.5( − 1) ̇ ) By substituting (42) into (28), i.e., replacing on the right- (52) hand side of (28) by the right-hand side of (42), one can obtain Comparing (52) with (24), we know that the term ()= ∫ − + ( − 1) d. (43) ( − ̇ ) in (52) represents the adiabatic process inside the cavern and the terms in the second line of (52) are By solving the integral equation (43), one can obtain associated with the heat transfer between the cavern air and the cavern wall. Equation (52) can be further reformulated into the ()= ( − )+ ( − 1) . (44) following bilinear form: ()= + ̇ + + ̇ + Adding (42) and (44) together gives ̇ ̇ + ̇ + ̇ + + + ()= − ( − 1) + ( − ) + ( (53) ̇ where - are constant coefficients as given in the − 1) . (45) Appendix. Equation (53) involves five variables, i.e., , ̇ , ( ) , , and and the first seven terms are bilinear terms. Equation (45) represents the change in temperature during the discharging process as a function of time and discharging E. Idle Process Considering Heat Transfer mass flow rate ̇ , where both the adiabatic process and the When idling, i.e., when neither charging nor discharging, heat transfer process are considered. Equation (45) can be heat transfer occurs between the cavern air and the cavern wall rewritten as if a temperature difference exists. By solving the integral equation (28), the change of air temperature in the cavern in the ( ) ()= − ( − 1) ̇ + − idle process can be expressed as . . ℎ + ℎ ̇ + ( − 1) ℎ ̇ + ℎ ̇ (46) ()= ( − ) + (54) . . Similarly, ̇ and ̇ can be linearized using where is the initial air temperature of the cavern in the idle process. According to the ideal gas law, one can obtain . . ̇ = ̇ + (̇ − ̇ ) (47) ( ) = ()/ (55) . . ( ) ̇ = ̇ + ̇ − ̇ (48) which can be expanded into (56) by substituting (54) therein: . . . . ̇ ̇ ̇ ̇ where = , = , and the ̇ ̇ ̇ ̇ ()= ( − ) / + / (56) setting of ̇ and ̇ will be given at the beginning of Section IV. Now, (46) can be reformulated into a bilinear form: Equation (56) can be reformed as ()= + ̇ + c ̇ + + ( ) = + (1 − )/ (57) (49) where - are constant coefficients as given in the The in (57) can be expressed as , which Appendix. Equation (49) involves four variables, i.e., , ̇ , can be linearized as follows: (), and , and the first three terms in the equation are bilinear terms. According to the ideal gas law, one can obtain = + ( − ) (58) , , ()= ( − ̇ ) ()/ (50) where = . By substituting (58) into (54) and (57), which can be expanded as follows by substituting (45) therein: one can obtain https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 8 ≤ ≤ , ∀ ∈ Ω (66) ()= ( − )( + ( − )/ )+ (59) + ≤ 1, ∀ ∈ Ω (67) ()= ( + ( − )/ ) ≤ ≤ , ∀ ∈ Ω (68) + 1 − − ( − ) (60) ≤ ≤ , ∀ ∈ Ω (69) Equations (59) and (60) can be reformulated into the following bilinear forms: where the values of the coefficients and are adopted from [26]. ( ) = + + + (61) The bilinear cavern model presented in the previous section, ()= + + + (62) i.e., (37), (41), (49), (53), (61) and (62), is re-written below where - are constant coefficients as given in the using superscript and (+1) to represent the indices of time Appendix. Equation (61) involves three variables, i.e., , steps. nd (), and , and the 2 term is the only bilinear term. ( ) ( ) ( ) ( ) Equation (62) involves three variables, i.e., , (), and , = + ̇ + ̇ + c ̇ + nd rd and the 2 and 3 terms are bilinear. + c + , ∀ ∈ Ω (70) This completes the deduction of the bilinear cavern model. In summary, the bilinear cavern model includes (37) and (41) ( ) ( ) ( ) ( ) = + ̇ + c ̇ + ̇ + for the charging process, (49) and (53) for the discharging , process, and (61) and (62) for the idle process. + + , ∀ ∈ Ω (71) ( ) ( ) ( ) = + ̇ + c ̇ + + , III. SELF-SCHEDULING OF COMPRESSED AIR ENERGY ∀ ∈ Ω (72) STORAGE IN THE DAY-AHEAD ELECTRICITY MARKET ( ) ( ) ( ) A. Self-scheduling Problem Using the Bilinear Cavern Model = + ̇ + + ̇ + ( ) ( ) ( ) ( ) In the SSP of CAES, the day-ahead electricity market is ̇ ̇ + ̇ + ̇ + + considered and the CAES plant is assumed to be a price taker + , ∀ ∈ Ω (73) (i.e., the CAES plant does not affect the electricity market price). ( ) The objective of SSP is to maximize the arbitrage revenue = + + + , ∀ ∈ Ω (74) obtained from selling electricity to and buying electricity from ( ) the day-ahead electricity market, i.e., selling electricity (in the = + + + , ∀ ∈ Ω discharging process) to the market in high-electricity-price (75) periods and buying electricity (in the charging process) from ( ) ( ) ( ) the market in lower-electricity-price periods. This implies that The ( , ) represents the temperature of the , , , the aim of SSP is to decide when to charge and discharge cavern air if the th step is a charging (discharging, idle) process. subjected to physical constraints of the CAES plant, including ( ) The temperature at the beginning of the (+1)th step, i.e., , the minimum/maximum pressure range of the cavern air. ( ) ( ) ( ) is equal to one of , , and according to the The objective function of the SSP can be expressed by (63), , , , ( ) ( ) which represents the net profit obtained by the CAES plant values of and , which is modelled in (76). The from the electricity market. Note that the operational costs of relationship between the pressure at the beginning of the ( ) ( ) ( ) ( ) charging and discharging power in the CAES are considered in (+1)th step, i.e., , and , , and can be , , , (63). Equation (64) represents the relationship between the similarly modeled as (77). The relationship among the mass of mass flow rate in (̇ ) and charging power ( ) and (65) ( ) air at two consecutive steps ( and ) and the represents the relationship between the mass flow rate out (̇ ) ( ) ( ) charging/discharging status indicators ( and ) can and discharging power ( ) [12]. Equation (66) represents the be expressed as (78). lower and upper bounds of the air pressure in the cavern. Equation (67) ensures that the charging and discharging ( ) ( ) ( ) ( ) ( ) ( ) = + + 1 − − , , processes do not occur at the same time. Equations (68) and (69) ( ) ( ) describe the ranges of charging and discharging power and their , ∀ ∈ Ω (76) relationships with the charging and discharging status ( ) ( ) ( ) ( ) ( ) ( ) indicators, i.e., and , respectively. = + + (1 − − , , ( ) ( ) maximize: = ∑ ( − ) − ( + ∈Ω ) , ∀ ∈ Ω (77) ) (63) ( ) ( ) ( ) ( ) ( ) = + ̇ Δ − ̇ Δ, ∀ ∈ ̇ = , ∀ ∈ Ω (64) Ω (78) The SSP of CAES using the bilinear model, i.e., (63)-(78), is ̇ = , ∀ ∈ Ω (65) https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 9 referred to as Model 1, which is an MIBLP model. The bilinear terms in Model 1 can be expressed as , which is equal to ( ) ( ) ( ) ( ) + /4 − − /4 . Then, + and − can be piecewise linearized. The detailed procedure of this linearization is available in Section IV-C of [40]. The resulting MILP model is referred to as Model 2. B. Flowcharts for Using the Bilinear Cavern Model The flowcharts for using the bilinear model, i.e., (70)-(78), to calculate the temperature, pressure, and mass of air in the charging, discharging, and idle processes are given in Figs. 4a, 4b, and 4c, respectively. The calculation procedures given in Figs. 4a, 4b, and 4c are recursive but consume very little time because equations (70)-(75) each only have one variable per step that can simply be calculated from the other known values. For example, is the only variable in (70) and all the other notations in (70) are known. Flowcharts given in Figs. 4a, 4b, (c) (d) and 4c are used in Sections IV-A to IV-C. Fig. 4. Flowchart for calculating the temperature, pressure, and mass of air using the bilinear model: (a) in the charging process, (b) in the discharging The flowchart for using the bilinear cavern model for an process, (c) in the idle process; (d) flowchart for using the proposed bilinear optimization problem is given in Fig. 4d, i.e., using (70)-(78) in model in an optimization model. an optimization problem. The bilinear cavern model part, i.e., (70)-(78), in an optimization model has 11 continuous C. Self-scheduling Problem Using Nonlinear and Constant- variables (i.e., , , and for = 1,2, ⋯ , and , temperature Cavern Models , , , , , ̇ , and ̇ for = , , , , , For the sake of completeness, the nonlinear cavern model 0,1,2, ⋯ , ( − 1)) and 2 binary variables (i.e., and given in [29] is rewritten here, i.e., (79)-(82). The SSP of CAES for = 0,1,2, ⋯ , ( − 1)). The flowchart given in Fig. using the nonlinear cavern model given in [29] consists of (63)- 4d is used in Sections IV-D and IV-E. Equations (70)-(78) (69), (76), and (78)-(82), and is referred to as Model 3. together with the objective and other constraints, e.g., (63)-(69), ( ) ( ) form an optimization problem that can be solved by an MINLP ( ) = + − ( ) solver, further details of which are given in Section IV-D. ( ) (79) ( ) Input , , , and at Input , , , and at ( ) time step time step ( ) = + − , ( ) Calculate and via Calculate and via (80) ( ) (70) and (71), respectively (72) and (73), respectively , , and ( + ) , , and ( ) ( ) = ( − ) + (81) are respectively used as , are respectively used as , , and for the next , and for the next ( ) ( ) ( ) = / (82) step in the charging process step in the charging process For the sake of comparison, the constant-temperature cavern model is also provided here, i.e., (83). The SSP of CAES using the constant-temperature model consists of (63)-(69), (76), and end ? end ? (83), which is referred to as Model 4. Models 1-4 will be compared in Section IV-E. (a) (b) ( ) = (83) IV. SIMULATION In this paper, the parameters for the Huntorf CAES plant are used for the calculations. The Huntorf CAES plant features two caverns with volumes of 141,000 and 169,000 m , respectively. Note that the maximum mass flow rate in the charging process (̇ ) is 108 kg/s for the whole plant and 49.12 kg/s for the first cavern, which is calculated from 108× . Similarly, the maximum mass flow rate in the discharging process (̇ ) https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 10 is 417 kg/s for the whole plant and 189.67 kg/s for the first the inaccuracy of the charging/discharging parts of the bilinear model, i.e., (70)-(73), is around 0.11%, which is quite small. cavern, which is calculated from 417× . In this This verifies the accuracy of the bilinear model for the given paper, the first cavern is used for calculations. The ̇ and initial temperature/pressure and mass flow rate. ̇ used in (35) and (36) are set to 49.12 and 22.74 kg/s, respectively. The ̇ and ̇ used in (47) and (48) are set TABLE II to 189.67 and 90.97 kg/s, respectively. The temperature at the THE MAPE BETWEEN THE RESULTS OBTAINED BY THE BILINEAR MODEL AND outlet of high-pressure compressor is high (about 230 °C), but THE ANALYTICAL MODEL GIVEN IN [29] IN EACH OF THE THREE PROCESSES. Charging process Discharging process Idle process drops to 50 °C after going through an aftercooler that is located Pressure 0.0011 0.0011 1.12× 10 between the high-pressure compressor and the cavern, i.e., the Temperature 0.0011 0.0011 1.12× 10 temperature of air injected into the caverns is equal to 50 °C. The other parameters for the Huntorf CAES plant are given in 70 50 Table I [28], [29]. TABLE I PARAMETERS FOR THE HUNTORF CAES PLANT. 25,000 m 718.3 J/(kg K) 1.4 66 bar 3 \$/MWh 286.7 J/(kg K) 141,000 m 40 °C 50 °C 3 \$/MWh Bilinear model Bilinear model Analytical model Analytical model Three specific processes are defined as follows and used to 45 20 0 4 8 12 16 0 4 8 12 16 verify the accuracy of the proposed bilinear model: Hour Hour  Charging process: Set the initial pressure (temperature) of (a) (b) the air in the cavern to 46 bar (20 °C). Charge the first Fig. 5. Results obtained by the proposed bilinear model and the analytical model cavern (141,000 m ) continuously for 16 hours at the in [29] during the charging process: a) pressure, b) temperature. maximum mass flow rate, i.e., ̇ =49.12 kg/s. 70 40  Discharging process: Set the initial pressure (temperature) Bilinear model Bilinear model Analytical model Analytical model of the air in the cavern to 66 bar (40 °C). Discharge the cavern continuously for 4 hours at the maximum mass flow rate, i.e., ̇ =189.67 kg/s.  Idle process: Set the initial pressure (temperature) of the air in the cavern to 60 bar (45 °C). Let the cavern be in the idle process for 16 hours. Reference [29] compares several existing CAES models with the measured data from the Huntorf CAES plant. The analytical 45 20 0 1 2 3 4 0 1 2 3 4 model in [29] is accurate and simpler than other existing Hour Hour analytical models. Thus, the analytical model in [29] is used as (a) (b) a benchmark model in this section to verify the accuracy of the Fig. 6. Results obtained by the proposed bilinear model and the analytical model proposed bilinear cavern model. in [29] during the discharging process: a) pressure, b) temperature. A. Model Verification - Comparison With Accurate Model In this subsection, the time interval is set to 1 second, i.e., Δ is equal to 1 second in (70)-(75). The pressure and temperature for each time interval of the charging (discharging, idle) process obtained from both the proposed bilinear model and the analytical model in [29] are plotted in Fig. 5 (Fig. 6, Fig. 7). Note that the difference between the results of the two models would be difficult to observe if the results for each second were shown in Figs. 5-7; therefore, the results for every 1000 (250, 1000) seconds are shown in Fig. 5 (Fig. 6, Fig. 7). Figs. 5-7 show that the pressure/temperature results obtained from the (a) (b) proposed bilinear model and the analytical model are quite Fig. 7. Results obtained by the proposed bilinear model and the analytical model close to one another. in [29] during the idle process: a) pressure, b) temperature. The mean absolute percentage errors (MAPEs) between the To comprehensively evaluate the accuracy of the bilinear results, in terms of both pressure and temperature, obtained model, different settings of initial temperatures/pressures and from the bilinear model and the analytical model during the mass flow rates are used in the charging/discharging/idle charging, discharging, and idle processes are tabulated in Table process. Different settings for the charging (discharging, idle) II. The last column of Table II shows that the idle part of the process are shown in rows C1-C6 (D1-D7, I1-I4) of Table III. bilinear model, i.e., (74) and (75), is almost as accurate as the nd rd When the pressure is larger (smaller) than 46 bar, the CAES is analytical model. The 2 and 3 columns of Table II show that Pressure (bar) Temperature ( C) https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 11 in normal (emergency) operating mode. Fig. 5 (Fig. 6) shows For each setting, the MAPE and mean absolute error (MAE) that the cavern temperature is relatively high (low) when the between the results (including both pressure and temperature) pressure is high (low). Therefore, the initial temperatures are set obtained by the bilinear and the analytical models is given in to 20 and 35 °C (50 and 35 °C) in the different settings for the the last four columns of Table III. The largest error in different charging (discharging) process as its initial pressure is settings for the charging process occurs at C2 with low initial relatively low (high). Both large and small mass flow rates are temperature and low mass flow rate; the MAPE and MAE for investigated in the normal operating mode as given in C1-C4 pressure (temperature) are 0.77% and 0.38 bar (0.77% and and D1-D4 where the large and small values are set to the 2.38 °C), respectively. The largest error in different settings for maximum mass flow rate and one tenth of it, respectively. To the discharging process occurs at D5 with high initial avoid a lengthy table, 1) only results that tend to have a large temperature and low mass flow rate; the MAPE and MAE for error in the emergency model are listed in Table III, i.e., low pressure (temperature) are 0.82% and 0.36 bar (0.82% and mass flow rate (associated with low charging power) and low 2.58 °C), respectively. Note that the MAPE and MAE are initial temperature as indicated by setting C1-C4, and low mass usually less than the worst-case values given above. The errors flow rate (associated with low discharging power) and high in different settings for the idle process are very small; the initial temperature as indicated by settings D1-D4; and 2) not MAPE for both pressure and temperature is less than 5E-5. This all the results of different settings for the idle process are listed indicates that the error between the proposed bilinear cavern because all of the different settings have very small errors, i.e., model and the accurate analytical model is small under a variety the MAPE is less than 5E-5. Note that ‘E-n’ and ‘E+n’ are used of circumstances. in this paper to represent ‘× 10 ’ and ‘× 10 ’, respectively. TABLE III DIFFERENT SETTINGS OF INITIAL PRESSURES AND TEMPERATURES AND CHARGING/DISCHARGING POWER AND THE CORRESPONDING MAPE AND MAE BETWEEN RESULTS OBTAINED BY THE BILINEAR CAVERN MODEL AND THE ANALYTICAL MODEL GIVEN IN [29] IN EACH OF THE THREE PROCESSES Pressure Temperature Setting Initial pressure (bar) Initial temperature (°C) Mass flow rate (kg/s) MAPE MAE (bar) MAPE MAE (°C) C1 46 20 49.12 0.0011 0.07 0.0011 0.35 C2 46 20 4.912 0.0077 0.38 0.0077 2.38 C3 46 35 49.12 0.0021 0.11 0.0020 0.64 C4 46 35 4.912 0.0017 0.08 0.0017 0.53 C5 30 20 4.912 0.0060 0.19 0.0060 1.84 C6 5 20 4.912 0.0047 0.03 0.0047 1.46 D1 66 50 189.67 0.0018 0.10 0.0018 0.57 D2 66 50 18.967 0.0078 0.50 0.0078 2.47 D3 66 35 189.67 0.0008 0.04 0.0008 0.24 D4 66 35 18.967 0.0056 0.37 0.0056 1.75 D5 46 50 18.967 0.0082 0.36 0.0082 2.58 D6 30 50 18.967 0.0071 0.20 0.0071 2.22 D7 5 50 18.967 0.012 0.047 0.012 3.82 I1 46 20 ---- 3.9E-5 1.8E-3 3.9E-5 1.2E-2 I2 5 20 ---- 4.2E-6 2.2E-5 4.2E-6 1.3E-3 I3 66 50 ---- 2.4E-5 1.6E-3 2.4E-5 7.7E-3 I4 5 50 ---- 1.8E-6 9.1E-6 1.8E-6 5.9E-4 Kelvin when calculating the MAPE (Kelvin is the unit of B. Model Verification - Comparison With Field-Measured temperature for all calculations involved in the bilinear cavern Data model). Note that the cavern pressure being higher (lower) than To further verify the accuracy of the proposed bilinear model, 46 bar indicates normal (emergency) operating mode; the CAES usually runs in normal operating mode. Therefore, the field-measured pressure and temperature data during a discharging process from [28] and a 10-stage proposed bilinear model is quite accurate when compared to field-measured data I in the normal operating mode of CAES. charging/discharging/idle process from [29] are used and referred to as field-measured data I and II, respectively. The For field-measured data II, the air mass flow rates in the 10 stages are given in Fig. 9a where negative (positive) values pressure and temperature results from both the proposed bilinear model and field-measured data I are plotted in Figs. 8b represent the discharging (charging) process and the zero value represents the idle process. In Fig. 9a, the 10 stages are divided and 8c, respectively. These figures show the pressures/temperatures obtained from the proposed bilinear by nine vertical dashed lines. In stage 1, the air mass flow rate linearly decreases from -125 to -175 kg/s and then increases to model are close to field-measured data I. Figs. 8b and 8c are both divided by a dashed line into two parts: the left-hand -125 kg/s. In stage 5, the CAES is discharging at a fixed air mass flow rate of -175 kg/s. In stages 3, 7, and 9, the CAES is (right-hand) part corresponds to pressure being higher (lower) than 46 bar. For the left-hand (right-hand) part, the MAPE and charging at a fixed air mass flow rate of 50 kg/s. In stages 2, 4, 6, 8, and 10, the CAES is in the idle stage, i.e., the air mass flow MAE between the results obtained by the bilinear model and field-measured data I are 0.47% and 0.27 bar (5.85% and 1.94 rate is zero. The pressures (temperatures) at each stage obtained from the bilinear model, analytical model, and field-measured bar) for the pressure and 0.19% and 0.56 °C (0.23% and 0.64 °C) for the temperature, where the unit of temperature is set to data II are plotted in Fig. 9b (Fig. 9c). Fig. 9b shows the https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 12 pressures obtained from both the bilinear model and the analytical model are quite close to field-measured data II except for stage 2. The MAPE and MAE between the pressure obtained by the bilinear (analytical) model and field-measured data II are (b) 1.01% and 0.54 bar (0.97% and 0.51 bar), respectively. In Fig. 9c, the temperature obtained from the bilinear model is closer to field-measured data II than the analytical model. The MAPE and MAE between the temperature obtained by the bilinear (analytical) model and field-measured data II are 3.16% and 1.18 °C (3.88% and 1.47 °C), respectively. In summary, the pressure (temperature) result obtained from the bilinear cavern model can well match the pressure (temperature) from both field-measured data I and II with a relative error of 0.47 and 1.01% (0.19 and 3.16%), respectively, in the normal operating mode of CAES. This comparison with (c) both an accurate model and two sets of field-measured data verifies the accuracy of the proposed bilinear cavern model. Bilinear model Analytical model Field-measured data II 0 5 10 15 20 25 Hour (a) Fig. 9. Mass flow rate into cavern (a) and comparison between the results obtained by the bilinear cavern model, analytical cavern model, and field- measured data II obtained from [29] for pressure (b) and temperature (c). C. Impact of Heat Transfer and Temperature To observe the impact of heat transfer, the results obtained (b) from the bilinear model with and without considering heat transfer in the charging (discharging) process are plotted in Fig. 10 (Fig. 11). In Fig. 10b, the slope of the left-hand part and the right-hand part of the dotted line (results of the model with heat transfer) is higher and lower, respectively, than the circled line (results of the model without heat transfer). The reason is that (c) when the cavern air temperature is lower (higher) than the cavern wall temperature, i.e., 40 °C, the heat transfer provides heat to (absorbs heat from) the cavern air and therefore the temperature obtained by the model with heat transfer increases faster (slower) than the model without heat transfer. According to the ideal gas law, i.e., = , for the same , , and , Fig. 8. Mass flow rate out of cavern (a) and comparison between the results higher (lower) temperature corresponds to higher (lower) obtained by the bilinear cavern model and field-measured data I from [28] for pressure (b) and temperature (c). pressure and therefore the relationship between the dotted and circled lines in Fig. 10a is similar to Fig. 10b. In Fig. 11, the dotted line has a smaller slope than the circled line. The reason is that the heat transfer provides heat to the cavern air and therefore the temperature decrease associated with the model with heat transfer is smaller than the model without heat transfer. Therefore, heat transfer clearly has a significant impact (a) on the temperature and pressure variation and, consequently, it is important to consider heat transfer in the cavern model. The results obtained from the constant-temperature model in the charging (discharging) process are also plotted in Fig. 10 (Fig. 11). Obviously, the pressure and temperature obtained from the constant-temperature model are quite different from those obtained with the bilinear model. Considering that the accuracy of the bilinear model was verified in Section V-A, Figs. 10 and 11 indicate that the constant-temperature cavern model is inaccurate. Therefore, it is necessary to use an accurate cavern model instead of the constant-temperature cavern model, especially in a real application problem. Pressure (bar) https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 13 Figs. 12a-12d show that the accuracy of the bilinear model in the charging and discharging processes decreases as the time interval increases. This is because the charging and discharging parts of the model utilize the assumption that the air injected into or released from the cavern in each time step is much smaller than the total amount of air in the cavern (i.e., ̇ Δ ≪ and ̇ Δ ≪ ); this assumption introduces higher error when the time interval is larger. Figs. 12e-12f show that the accuracy of the bilinear model in the idle process does not change with the time interval; this is because the idle part of the model does not utilize the above assumption. Tables IV-V show that the error and relative error of the temperature and pressure in both the charging and discharging processes are small when the time interval is less than or equal to 10 minutes. When the time interval is equal to 20 minutes, (a) (b) the errors (relative errors) of the pressure in the charging and Fig. 10. Results obtained from the bilinear cavern model (with and without discharging processes are 0.121 and 0.045 bar (0.17 and 0.09%), considering heat transfer) and the constant-temperature cavern model in the respectively, which are still relatively small. When the time charging process: a) pressure of the air in the cavern and b) temperature of the air in the cavern. interval is equal to 60 minutes, the error (relative error) in both the charging and discharging processes is relatively large. Table VI shows that the idle process of the model is quite accurate for all time intervals. Therefore, Fig. 12 and Tables IV-VI show that the accuracy of the bilinear cavern model is high, moderate, and relatively low when the time interval is between 1 second and 10 minutes, equal to 20 minutes, and equal to 60 minutes, respectively. (a) (b) Fig. 11. Results obtained from the bilinear cavern model (with and without considering heat transfer) and the constant-temperature cavern model in the discharging process: (a) pressure of the air in the cavern and (b) temperature of (a) (b) the air in the cavern. D. Different Time Intervals In power system operation problems, the time interval is usually longer than one second, e.g., the time intervals of economic dispatch and unit commitment are both usually 1 hour, respectively. Therefore, it is necessary to know whether the proposed bilinear model is accurate for different time intervals. (c) (d) In this regard, the status of the air in the cavern is calculated using the procedure given in Figs. 4a, 4b, and 4c for different time intervals (1 second, 1 minute, 5 minutes, 10 minutes, 20 minutes, and 60 minutes) using the same initial status. The final temperature and pressure of the charging, discharging, and idle processes obtained by the bilinear model and the analytical model are plotted in Fig. 12. The error and relative error (values given in round brackets) between the results, in terms of final temperature and pressure, obtained from the two models are (e) (f) nd th th th th shown in Table IV, where the 2 -5 , the 6 -9 , and the 10 - Fig. 12. Final temperature/pressure obtained by both the analytical model and th 13 rows show the error/relative error in the charging, the bilinear model in different processes using different time intervals: (a) final discharging, and idle processes, respectively. Again, the unit of temperature in the charging process, (b) final pressure in the charging process, (c) final temperature in the discharging process, (d) final pressure in the temperature is set to Kelvin when calculating the relative error discharging process, (e) final temperature in the idle process, and (f) final of temperature. pressure in the idle process. Pressure (bar) Pressure (bar) o Temperature ( C) Temperature ( C) o o o Temperature ( C) Temperature ( C) Temperature ( C) Pressure (bar) Pressure (bar) Pressure (bar) https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 14 TABLE IV value represents a better solution as the goal of SSP is to ERROR (RELATIVE ERROR IN ROUND BRACKETS) BETWEEN THE SOLUTION maximize the profits. When the numbers of time steps are 10, OBTAINED BY THE BILINEAR MODEL AND THE ANALYTICAL MODEL IN THE 24, and 72, the objectives of Model 2 are 1.3, 7.4, and 3.6% CHARGING PROCESS USING DIFFERENT TIME INTERVALS higher than Model 3 and 1.3, 7.4, and 1.1% higher than Model Interval 1 s 1 min 5 min 10 min 20 min 60 min 1, respectively. 0.066 0.051 -0.010 -0.087 -0.241 -0.872 Temperat The electricity market price, power output, and pressure in ure (°C) (0.02%) (0.02%) (-3E-5) (-0.03%) (-0.08%) (-0.27%) the 24-time-step case are plotted in Fig. 13. Note that the power 0.002 0.007 0.031 0.061 0.121 0.356 Pressure( output of the CAES in Fig. 13b equals the discharging power bar) (0.002%) (0.01%) (0.05%) (0.09%) (0.17%) (0.52%) minus the charging power. The high price hours are 9-12 and 18-21. Fig. 13b indicates that the CAES generally discharges TABLE V power at high price hours and charges power at low price hours ERROR (RELATIVE ERROR IN ROUND BRACKETS) BETWEEN THE SOLUTION to maximize the profit. The difference between the results of OBTAINED BY THE BILINEAR MODEL AND THE ANALYTICAL MODEL IN THE Models 1-3 lies in the power output in hours 1, 12, 18, and 20- DISCHARGING PROCESS USING DIFFERENT TIME INTERVALS 22. A detailed comparison between Models 2 and 3 is given Interval 1 s 1 min 5 min 10 min 20 min 60 min below. A similar analysis can be applied to Models 2 and 1 but -0.386 -0.350 -0.201 -0.014 0.366 1.968 Temperat is not provided. ure (°C) (-0.13%) (-0.12%) (-0.07%) (-5E-5) (0.12%) (0.67%) The results of Model 2 discharge more power at hours 12 and -0.060 -0.055 -0.034 -0.008 0.045 0.270 Pressure 21 and charge more power at hour 20 compared to the results (bar) (-0.13%) (-0.12%) (-0.07%) (-0.02%) (0.09%) (0.58%) of Model 3, as detailed in Table VIII. In Table VIII, the last column provides the total profit in hours 12, 20, and 21 and TABLE VI clearly shows that Model 2 achieves higher profits than Model ERROR (RELATIVE ERROR IN ROUND BRACKETS) BETWEEN THE SOLUTION 3 and coincides with Table VII. OBTAINED BY THE BILINEAR MODEL AND THE ANALYTICAL MODEL IN THE IDLE PROCESS USING DIFFERENT TIME INTERVALS Linearizing the bilinear model into an MILP problem results Interval 1 s 1 min 5 min 10 min 20 min 60 min in high accuracy as shown in [40], i.e., Model 2 is almost as -2E-4 -2E-4 -2E-4 -2E-4 -2E-4 -2E-4 accurate as Model 1. Therefore, the higher profit obtained from Temperat ure (°C) Model 2 indicates that a better solution has been obtained from (-1E-6) (-1E-6) (-1E-6) (-1E-6) (-1E-6) (-1E-6) Model 2 than Models 1 and 3 and the solutions obtained from -1E-4 -1E-4 -1E-4 -1E-4 -1E-4 -1E-4 Pressure( Models 1 and 3 are not optimal. This is because MILP problems bar) (-2E-6) (-2E-6) (-2E-6) (-2E-6) (-2E-6) (-2E-6) are easier to solve than MINLP or MIBLP problems. This shows the advantage of the proposed bilinear cavern model, E. Self-scheduling Problem of Compressed Air Energy which can be converted to an MILP problem and is therefore Storage in Day-ahead Electricity Market suitable for integration into optimization problems considering This subsection presents the results of solving the SSP of CAES. CAES using different cavern models. When the time interval is set to 60 minutes, the day-ahead SSP has 24 steps, which TABLE VII introduces a relatively small computational burden. However, OBJECTIVE VALUES (\$) OF THE SELF-SCHEDULING PROBLEM OF COMPRESSED the error is relatively large as shown in the previous subsection. AIR ENERGY STORAGE USING DIFFERENT CAVERN MODELS (NUMBERS IN When the time interval is set to 20 minutes, the error is BRACKETS ARE THE RELATIVE DIFFERENCE BETWEEN MODEL 2 AND MODEL (= 1,3)) relatively small and the day-ahead SSP has 72 steps. When the Number of time Model 3 - Model 1 - Model 2 - time interval is set to 10 minutes or smaller, the error decreases steps ( ) MINLP [29] MIBLP MILP but the computational burden increases. Therefore, the time 10 2.28E+5 (1.3%) 2.28E+5 (1.3%) 2.31E+5 interval in the SSP is set to 20 minutes as a trade-off between 24 2.51E+5 (7.4%) 2.51E+5 (7.4%) 2.71E+5 72 5.43E+5 (3.6%) 5.57E+5 (1.1%) 5.63E+5 the accuracy and computational burden. 1) Comparison Among Models 1-3 Models 1 and 3, described in Sections III-A and III-C, TABLE VIII respectively, are both solved by a popularly used MINLP solver, THE PRICE AND THE SOLUTION OBTAINED BY MODELS 2 AND 3 IN HOURS 12, 20, AND 21 OF THE SELF-SCHEDULING PROBLEM OF COMPRESSED AIR i.e., BARON [46], with the values of objective functions (i.e., ENERGY STORAGE rd nd the profits from the electricity market) given in the 3 and 2 Hour 12 Hour 20 Hour 21 Total profit (\$) columns of Table VII, respectively. Model 2 is solved by Price (\$) 225.58 240.26 229.65 -- th CPLEX with the objective function value given in the 4 Power output (MW) 154.49 40.0 40.0 -- Model 3 Profit (\$) 34386.4 9490.4 9066.0 52942.8 column of Table VII. Models 1-3 are MIBLP, MILP, and Power output (MW) 290.0 -60.0 101.71 -- MINLP models, respectively, as mentioned in Section III. Model 2 Profit (\$) 64548.2 -14595.6 23052.6 73005.2 These three models are solved using different numbers of time steps (i.e., ). The stopping criteria for both the BARON and CPLEX solvers are set to a maximum time of 8 hours and a gap of 0.1%; the solving process will terminate as soon as one of the two criteria is met. Table VII shows that the objective function value obtained from Model 2 is higher than that from Models 1 and 3 for all three different numbers of steps. Note that a larger objective https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 15 (a) (a) -100 0 5 10 15 20 25 Constant-temperature model Accurate value (b) (b) 0 5 10 15 20 25 70 Constant-temperature model (c) Accurate value (c) 0 5 10 15 20 25 Fig. 13. The electricity market price (a) and the solution obtained by Models 1- 3 in the 24-time-step case for the power output (b) and pressure (c). Hour (h) Fig. 14. The solution for the self-scheduling problem: (a) the power output 2) Comparison Between the Bilinear and Constant- obtained from the constant-temperature model, (b) the temperature obtained temperature Cavern Models from the constant-temperature model and the corresponding accurate value, and (c) the pressure obtained from the constant-temperature model and the To show the advantage of the proposed bilinear cavern model corresponding accurate value. over the constant-temperature model, both models are used to solve the SSP and comparative analysis is given as follows. The F. Cavern Efficiency results, including the power output, temperature, and pressure, The efficiency of the cavern is analyzed in this subsection. are given in Fig. 14. In Figs. 14b and 14c, the dots represent the The internal energy is calculated from , i.e., the product of temperature and pressure obtained from the SSP using the mass of air , a constant specific heat value , and temperature constant-temperature cavern model as detailed in Section III-C. . Let the cavern efficiency be the ratio of the internal energy Given the power output obtained from the SSP using the of all the air discharged out of the cavern in a period of time to constant-temperature cavern model as input, the SSP using the the internal energy of all the air charged into the cavern in the bilinear cavern model is used to calculate the temperature and same period of time. Note that the temperature of the air pressure as shown in the circles in Figs. 14b and 14c. That is, charged into the cavern is always equal to 50 °C because of the the circles given in Figs. 14b and 14c are the accurate values for use of aftercoolers. the charging/discharging process given in Fig. 14a. Fig. 14b The initial temperatures of cavern air in the first day are set shows that the temperature during the charging/discharging to 35, 40, and 45 °C, respectively. For each setting, Model 3 for process varies in a range of about 30 K, which is quite different the SSP of CAES in one day is solved; the internal energies of from constant. In Fig. 14c, the pressure obtained from the air injected into and released out of the cavern on this day are constant-temperature cavern model varies from 46 to 56.43 bar nd th given in the 2 and 4 columns of Table IX, respectively; and (shown in dots); however, the pressure actually varies from the heat transferred from the cavern wall to the cavern air on 40.68 to 63.44 bar (shown in circles). At both hours 21 and 22, rd this day is given in the 3 column of Table IX. Note that the the pressure obtained from the constant-temperature cavern rd positive (negative) values in the 3 column of Table IX indicate model is 46 bar; however, the pressure actually drops down to that less (more) heat is transferred to the cavern wall than is 40.68 and 40.79 bar, respectively, at these two time points. That received from the cavern wall. is, the pressure obtained from the constant-temperature cavern The cavern efficiency for each setting is given in the last model is quite inaccurate and the actual pressure drops far column of Table IX. Table IX shows that the cavern efficiency below the minimum bound of the normal operating range (i.e., is higher than 1 when the initial temperature is 35 °C. This is 46 bar as shown in the dashed line in Fig. 14c). This inaccurate because the temperature of the cavern wall (40 °C) is higher constant-temperature cavern model, however, has been adopted than 35 °C and thus heat is transferred to the cavern air from the in a variety of power system application areas, e.g., [31], [32], cavern wall. That is, the cavern air gains energy from the cavern [33], [34]. The proposed bilinear cavern model can replace the wall and therefore the cavern efficiency is higher than 1. But constant-temperature cavern model and is suitable for these when the initial temperature is 40 or 45 °C, the cavern power system applications and beyond, which shows the value efficiency is lower than 1; this is because the temperature of of the proposed bilinear cavern model. cavern air is equal to or higher than that of the cavern wall and therefore heat is transferred to rather than received from the https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 16 cavern wall. That is, the cavern air loses energy to the cavern compressed air energy storage. The accuracy of the proposed wall and therefore the cavern efficiency is lower than 1. bilinear model decreases as the time interval increases. For time To determine the average cavern efficiency for a longer intervals between 1 second and 10 minutes, equal to 20 minutes, period of time, the efficiencies for 10 consecutive days for each and 60 minutes or longer, the bilinear cavern model has high, nd th setting are plotted in Fig. 15. For each of the 2 to 10 days, moderate, and relatively low accuracy, respectively. Simulation the initial temperature of the cavern air in the next day is set to results also show that heat transfer has an obvious and be the temperature of the cavern air at the end of the previous measurable effect on the variation of temperature and pressure day and then the SSP of CAES in the day is solved to calculate of the air in the cavern. Therefore, it is necessary to consider the cavern efficiency for that day. The figure shows that the heat transfer in the cavern model. The constant-temperature cavern efficiencies on the first day are quite different, but are cavern model is also shown to be inaccurate. nd th close to one another for the 2 -10 days. The average The self-scheduling problems of compressed air energy efficiency for all days is 0.9572 and the average efficiency for storage using different cavern models are solved. The all days excluding the first day is 0.9579. Therefore, the average simulation results show that the self-scheduling problem of cavern efficiency is about 0.96. compressed air energy storage using the proposed bilinear cavern model can be straightforwardly converted into a mixed- integer linear programming model that is easier to solve than TABLE IX the mixed-integer nonlinear programming. The self-scheduling INTERNAL ENERGY OF AIR CHARGED IN AND DISCHARGED OUT, HEAT TRANSFERRED FROM CAVERN WALL, AND THE CAVERN EFFICIENCY problem of compressed air energy storage using the constant- Initial Internal Heat Internal Cavern temperature cavern model can result in infeasible solutions. temperature energy of transferred energy of efficiency However, by properly setting the time interval, the proposed of cavern air air in (J) from cavern air out (J) bilinear cavern model is accurate and suitable for use in power (°C) wall (J) system optimization problems and is superior to the existing 35 3.45E+15 0.14E+15 3.59E+15 1.04 40 3.40E+15 -0.17E+15 3.23E+15 0.95 highly nonlinear cavern model and the constant-temperature 45 3.42E+15 -0.48E+15 2.94E+15 0.86 cavern model. APPENDIX = ( − 2) , = , = = , = ℎ ̇ − ℎ ̇ , = , ( ) = − 1 , = , ( ) = − 1 , = , = , = , = Δ ( ) , Fig. 15. Cavern efficiencies for 10 consecutive days using different initial temperatures for the first day. = / , = ( ) / , V. CONCLUSION ( ) = − 2 Δ− ℎ Δ − ℎ − ℎ , This paper proposed an accurate bilinear cavern model for = Δ( − 1)( ) − (ℎ + ℎ )( − 2), compressed air energy storage based on the ideal gas law and ( ) ( )( ) c = Δ + ℎ Δ − Δ − 1 − the first law of thermodynamics. The accuracy of the proposed ( )( ) bilinear cavern model is verified via comparison with both an ℎ + ℎ 3 − , accurate analytical model in the literature and two sets of field- . . = ℎ Δ ̇ − ℎ Δ̇ − ℎ Δ − ℎ ̇ + measured data. ℎ ̇ , Simulation results show that the mean absolute percentage c = − ( − 2), error (mean absolute error) between the bilinear cavern model and the accurate analytical model for pressure (temperature) are . = Δℎ ̇ − Δℎ ̇ + ℎ Δ − 3 − no more than 0.82% and 0.36 bar (0.82% and 2.58 °C) for a variety of conditions. Note that the errors are usually less than ( ) ( )( ) the worst-case values given above. Simulation results also show = − 1 Δ− ℎ − 0.5 − 2 ℎ + ℎ − that the pressure (temperature) results obtained from the Δ ℎ / − ℎ , bilinear cavern model can well match the pressure (temperature) c = ( Δ( ) + ℎ ), from the two sets of field-measured data considered, i.e., ( ( )( ) ) = Δ 1 − + ℎ Δ+ Δℎ , datasets I and II, with a relative error of 0.47 and 1.01% (0.19 and 3.16%), respectively, in the normal operating mode of = ℎ ̇ − ℎ ̇ + ℎ , Cavern efficiency https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 17 . [10] M. Budt, D. Wolf, R. Span, J. Yan, “A review on compressed air energy = − ℎ ̇ + ℎ − ℎ ̇ Δ/ + storage: Basic principles, past milestones and recent developments”, . . Applied Energy, vol. 170, pp. 250-268, 2016. ℎ ̇ − ̇ + 0.5 ( − 2)ℎ ̇ − [11] J. Wang, K. Lu, L. Ma, J. Wang, M. Dooner, S. Miao, J. Li, D. Wang, "Overview of Compressed Air Energy Storage and Technology ̇ , Development”, Energies, vol. 10, no. 7, pp. 1-22, 2017. = Δℎ ̇ − ̇ , [12] X. Luo, J. Wang, C. Krupke, Y. Wang, Y. Sheng, J. Li, Y. Xu, D. Wang, S. Miao, H. Chen, “Modelling study, efficiency analysis and optimisation ( ( ) ( ) ) = ℎ + ℎ − − 1 Δ− ℎ , of large-scale adiabatic compressed air energy storage systems with low- = ℎ , temperature thermal storage”, Applied Energy, vol. 162, pp. 589–600, . . c = ℎ ̇ − ℎ ̇ − ℎ + ℎ ̇ − [13] E. Barbour, D. Mignard, Y. Ding, Y. Li, “Adiabatic compressed air energy storage with packed bed thermal energy storage”, Applied Energy, vol. 155, ℎ ̇ , pp. 804-815, 2015. [14] H. Peng, R. Li, X. Ling, H. Dong, “Modeling on heat storage performance = ℎ + ℎ ̇ − ℎ ̇ , of compressed air in a packed bed system”, Applied Energy, vol. 160, pp. 1–9, 2015. c = − Δ+ , [15] N. Hartmann, O. Vöhringer, C. Kruck, L. Eltrop, “Simulation and analysis of different adiabatic compressed air energy storage plant configurations”, = − ℎ , Applied Energy, vol. 93, pp. 541-548, 2012. ( ( ) ) = Δℎ + ℎ + Δℎ + ℎ , [16] Y. Zhang, Y. Xu, X. Zhou, H. Guo, X. Zhang, H. Chen, “Compressed air = − Δℎ , energy storage system with variable configuration for accommodating large-amplitude wind power fluctuation”, Applied Energy, vol. 239, pp. = ℎ , 957-968, 2019. = − Δℎ + ℎ ̇ − ℎ ̇ , [17] J.D. Wojcik, J. Wang, “Feasibility study of Combined Cycle Gas Turbine (CCGT) power plant integration with Adiabatic Compressed Air Energy = ℎ + ℎ ̇ − ℎ ̇ , Storage (ACAES)”, Applied Energy, vol. 221, pp. 477-489, 2018. . [18] I. Ortega-Fernández, S. A. Zavattoni, J. Rodríguez-Aseguinolaza, B. = − ̇ − ̇ , D'Aguanno, M.C. Barbatob, “Analysis of an integrated packed bed thermal energy storage system for heat recovery in compressed air energy ( ) = Δℎ + ℎ ̇ − ̇ , storage technology”, Applied Energy, vol. 205, pp. 280-293, 2017. ( ) [19] X. Zhang, Y. Xu, X. Zhou, Y. Zhang, W. Li, Z. Zuo, H. Guo, Y. Huang, = , c = − , H. Chen, “A near-isothermal expander for isothermal compressed air energy storage system”, Applied Energy, vol. 225, pp. 955-964, 2018. ( ) c = − , = − − − 1 , [20] B. Bollinger, “Technology Performance Report - SustainX Smart Grid Program-Demonstration of Isothermal Compressed Air Energy Storage to ( ) = − , = 1 − + . Support Renewable Energy Production”, Apr. 1, 2015. Available:https://www.smartgrid.gov/project/sustainx_inc_isothermal_co mpressed_air_energy_storage.html. [21] A. Arabkoohsar, M. Dremark-Larsen, R. Lorentzen, G.B. Andresen, ACKNOWLEDGEMENTS “Subcooled compressed air energy storage system for coproduction of heat, The work was supported in part by the Natural Sciences and cooling and electricity”, Applied Energy, vol. 205, pp. 602-614, 2017. Engineering Research Council (NSERC) of Canada and the [22] B. Kantharaj, S. Garvey, A. Pimm “Compressed air energy storage with Saskatchewan Power Corporation (SaskPower). liquid air capacity extension”, Applied Energy, vol. 157, pp. 152–164, [23] S. Succar, D. C. Denkenberger, R.H. Williams, “Optimization of specific REFERENCES rating for wind turbine arrays coupled to compressed air energy storage”, Applied Energy, vol. 96, pp. 222-234, 2012. [1] Global wind energy council, Global Wind Energy Outlook 2016, [24] X. Luo, M. Dooner, W. He, J. Wang, Y. Li, D. Li, O. Kiselychnyk, Available: http://www.gwec.net. “Feasibility study of a simulation software tool development for dynamic [2] U. S. Department of Energy, “EPRI-DOE Handbook of Energy Storage modelling and transient control of adiabatic compressed air energy storage for Transmission & Distribution Applications: Final Report,” 2003. with its electrical power system applications”, Applied Energy, vol. 228, [3] X. Luo, J. Wang, M. Dooner, and J. Clarke, “Overview of current pp. 1198-1219, 2018. development in electrical energy storage technologies and the application [25] S. Briola, P. D. Marco, R. Gabbrielli, J. Riccardi, “A novel mathematical potential in power system operation”, Applied Energy, vol. 137, pp. 511- model for the performance assessment of diabatic compressed air energy 536, 2015. storage systems including the turbomachinery characteristic curves”, [4] International Energy Agency, “Technology Roadmap Energy Storage,” Applied Energy, vol. 178, pp. 758-772, 2016. [Online]. Available: https://www.iea.org, accessed on 19 March 2014. [26] T. Das and J.D. McCalley, “Compressed air energy storage”, Educational [5] World Energy Council, World Energy Resources – Hydropower 2016, Chapter, Iowa State Univ. 2012. [Online] https://www.worldenergy.org [27] R. Kushnir, A. Ullmann, A. Dayan, “Thermodynamic models for the [6] BBC Brown Boveri, Huntorf Air Storage Gas Turbine Power Plant, temperature and pressure variations within adiabatic caverns of Available: http://www.solarplan.org/Research/BBC_Huntorf_engl.pdfS. compressed air energy storage plants”, J. Energy Resources Technology, Succar and R. H. Williams. Compressed Air Energy Storage: Theory, vol. 134, no. 2, pp. 1-10, 2012. Resources, and Applications for Wind Power. Princeton, NJ: Princeton [28] M. Raju, and S.K. Khaitan, “Modeling and simulation of compressed air University. 2008. storage in caverns: A case study of the Huntorf plant”, Applied Energy, [7] U. S. Department of Energy / National Energy Technology Laboratory vol. 89, no. 1, pp. 474-481, 2012. (NETL), “Technical and Economic Analysis of Various Power Generation [29] C. Xia, Y. Zhou, S. Zhou, P. Zhang, F. Wang, “A simplified and unified Resources Coupled with CAES Systems: Final Report”, May 2011. analytical solution for temperature and pressure variations in compressed [8] Power South Energy Cooperative, Compressed air energy storage-- air energy storage caverns”, Renewable Energy, vol. 74, pp. 718-726, 2015. McIntosh Power Plant, Alabama, U. S. [Online]. http://www.powersouth. [30] W. He, X. Luo, D. Evans, J. Busby, S. Garvey, D. Parkes, J. Wanga, com/files/CAES%20Brochure%20[FINAL].pdf “Exergy storage of compressed air in cavern and cavern volume estimation [9] Rocky Mountain Power, “Alberta Saskatchewan Intertie Storage (ASISt)”, of the large-scale compressed air energy storage system”, Applied Energy, [Online] http://rockymountainpower.ca/ASISt.html vol. 208, pp. 745-757, 2017. [31] H. Khani, M.R.D. Zadeh, A.H. Hajimiragha, “Transmission congestion relief using privately owned large-scale energy storage systems in a https://doi.org/10.1016/j.apenergy.2019.03.104 accepted by Applied Energy on March 2019 18 competitive electricity market”, IEEE Trans. Power Syst., vol. 31, no. 2, pp. 1449-1458, 2016. [32] S. Shafiee, H. Zareipour, A.M. Knight, N. Amjady, B. Mohammadi- Ivatloo, “Risk-constrained bidding and offering strategy for a merchant compressed air energy storage plant”, IEEE Trans. Power Syst., vol. 32, no. 2, pp. 946-957, 2017. [33] H. Daneshi and A.K. Srivastava, “Security-constrained unit commitment with wind generation and compressed air energy storage”, IET Gener. Transm. Distrib., vol. 6, no. 2, pp. 167-175, 2012. [34] X. Luo, J. Wang, M. Dooner, J. Clarke, C. Krupke, “Overview of current development in compressed air energy storage technology”, Energy Procedia, vol. 62, pp. 603-611, 2014. [35] T.D. Eastop and A.C. McConkey. Applied Thermodynamics for Engineering Technologists. Delhi, India: Pearson Education, 2009. [36] Gilbert Strang, Calculus, Wellesley, MA, USA: Wellesley-Cambridge Press, 2010. [37] I. Grossmann, “Review of nonlinear mixed–integer and disjunctive programming techniques”, Optimization and Engineering, vol. 3, pp. 227- 252, 2002. [38] S. Sager, C.M. Barth, H. Diedam, M. Engelhart, J. Funke, “Optimization as an analysis tool for human complex problem solving”, SIAM J. Optim, vol. 21, no. 3, pp. 936-959, 2011. [39] P.M. Castro, “Tightening piecewise McCormick relaxations for bilinear problems”, Computers and Chemical Engineering, vol. 72, pp. 300-311, [40] A. Zare, C.Y. Chung, J.P. Zhan, S.O. Faried, “A distributionally robust chance-constrained MILP model for multistage distribution system planning with uncertain renewables and loads”, IEEE Trans. Power Systems, vol. 33, no. 5, pp. 5248-5262, 2018. [41] E. Johnson, G.L. Nemhauser, M. Savelsbergh, “Progress in linear programming based algorithms for integer programming: An exposition”, INFORMS J. Computing, vol. 12, no. 1, pp. 2-23, 2000. [42] IBM, IBM ILOG CPLEX V12.1: User's manual for CPLEX, http://www.ibm.com, 2017. [43] Gurobi Optimization, Inc., Gurobi Optimizer Reference Manual, http://www.gurobi.com, 2017. [44] H.Y. Yamin and S.M. Shahidehpour, “Self-scheduling and energy bidding in competitive electricity markets”, Electric Power Syst. Res., vol. 71, no. 3, pp. 203-209, 2004. [45] G. Pedrick, “Compressed air energy storage engineering and economic study - final report by New York state energy research and development authority”, 2009, [Online]. https://www.nyserda.ny.gov [46] N.V. Sahinidis, BARON 17.8.9: Global Optimization of Mixed-Integer Nonlinear Programs, User's manual, 2017. [47] Y. Zimmels, F. Kirzhner, B. Krasovitski, “Design criteria for compressed air storage in hard rock”, Energy & Environment, vol. 13, no. 6, pp. 851- 872, 2002.

### Journal

Electrical Engineering and Systems SciencearXiv (Cornell University)

Published: Mar 16, 2019

### There are no references for this article.

Access the full text.

Sign up today, get DeepDyve free for 14 days.