An artificial intelligence and mechanism model fused compressed air energy storage system rapid design method

Abstract

The application discloses a kind of artificial intelligence and mechanism model fusion compressed air energy storage system rapid design method.First, the thermodynamic model of compressed air energy storage system that can reflect physical mechanism-economic analysis coupled model is constructed, and multidimensional performance database is established;Second, gradient boosting decision tree is used as the basis to train artificial intelligence agent model, introduce Shapley Additive Explanations (SHAP) method, quantitatively reveal the influence of system configuration parameters and operating conditions on the core performance indicators of the system, and build an explicit correlation of comprehensive performance;Finally, the explicit correlation is embedded as the performance evaluation kernel into the multi-objective optimization and objective weighting decision process, to realize the rapid optimization of the key design parameters of compressed air energy storage system.Compared with the traditional method relying on numerical simulation or black box model, the application significantly reduces the computational cost under the premise of ensuring physical rationality and interpretability, and outputs design results with clear parameter contribution explanation and applicable domain boundary.

Description

Technical Field

[0001] This invention belongs to the field of compressed air energy storage system design and optimization technology, specifically relating to a rapid design method for compressed air energy storage systems oriented towards scheme comparison and parameter determination. This method is based on a coupled thermodynamic model and economic analysis model, combined with interpretable artificial intelligence and multi-objective optimization to achieve rapid optimization of key design parameters. Background Technology

[0002] Compressed air energy storage technology, with its advantages of large scale, long lifespan, and low cost, has become one of the key energy storage technologies supporting the stable operation of grids with a high proportion of renewable energy. For advanced adiabatic compressed air energy storage systems, represented by medium- and high-temperature thermal storage loops, engineers need to coordinate and balance multiple design variables during the project planning and preliminary design phases, such as the number of compression / expansion stages, the distribution of pressure ratios at each stage, the scale of thermal storage units, gas storage parameters, and operating conditions, to achieve a compromise and optimal balance among key performance indicators such as energy conversion efficiency, energy storage space utilization, and life-cycle economics within the feasible engineering domain. This process requires both reliable characterization of complex physical processes and consideration of computational efficiency and the breadth of scheme comparisons; its difficulty increases exponentially with the number of design dimensions and operating scenarios.

[0003] Existing design methods for compressed air energy storage systems can be mainly categorized into two types. The first type is based on a coupled thermodynamic model and economic analysis model. It involves modularly modeling the compressor, expander, heat exchanger, gas storage tank, and heat storage-regenerative loop to quantitatively calculate system-level performance indicators under given configurations and operating conditions. This type of method exhibits high physical reliability in performance evaluations under single or a small number of typical operating conditions. However, when conducting systematic sensitivity analysis, scheme scanning, and multi-objective optimization in a multi-dimensional parameter space, it requires repeatedly solving for thousands of candidate parameter combinations, accumulating computation time to several hours or even days. This makes it difficult to meet the dual requirements of depth of scheme exploration and computational efficiency in early-stage engineering design.

[0004] The second type of approach uses data-driven AI proxy models to replace the thermodynamic-economic analysis coupled models, enabling rapid prediction of system performance indicators. However, commonly used AI proxy models often have multiple nonlinear mapping structures, with complex and opaque internal decision-making logic, making them typical "black box" models. When the proxy model provides a recommended solution, engineers find it difficult to verify its physical rationality, nor can they solidify the optimization process into explicit knowledge that can be directly referenced in design specifications.

[0005] Furthermore, after obtaining the non-dominated solutions (Pareto optimal solution set) among various system-level performance indicators using multi-objective optimization algorithms, it is still necessary to weigh and prioritize among multiple competing objectives. Without a structured, reproducible, and objective multi-indicator decision-making method, the final decision-making process is difficult to quantify and verify, further hindering the standardization and normalization of the design process.

[0006] In summary, existing technologies often struggle to balance computational efficiency and model interpretability when addressing the complex multi-objective optimization problems in the early-stage engineering design of compressed air energy storage systems. Therefore, developing a novel design method and tool that deeply integrates physical insights and data intelligence, while possessing rapid evaluation capabilities, clearly defining physical relationships, and providing objective decision support, is an important direction for improving the efficiency of compressed air energy storage system engineering design. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings in the current design and optimization of compressed air energy storage systems by proposing a rapid design method for compressed air energy storage systems that integrates artificial intelligence and mechanistic models. This method enables rapid design of compressed air energy storage systems from scenario requirements to parameter schemes, significantly improving design efficiency and overall performance while ensuring physical rationality and interpretability.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A rapid design method for compressed air energy storage systems that integrates artificial intelligence and mechanistic models includes the following steps: Step 1): Construct a thermodynamic model-economic analysis coupled model of the compressed air energy storage system as a system-level mechanism model. Combine system configuration parameters and operating condition parameters to form input features, and use multiple system performance indicators that characterize energy conversion efficiency, energy storage space utilization rate and life cycle economy as output features. Step 2): Under the design boundary and engineering constraints, Latin hypercube sampling is used to generate sample points in the input parameter space, and the mechanism model is called to calculate the corresponding system-level performance index to construct a multi-dimensional performance database; Step 3): Construct a proxy model based on the performance database, and introduce the SHAP posterior interpretation method to decompose the feature contribution and characterize the dependency relationship of the proxy model, and express the model prediction relationship explicitly as a comprehensive performance explicit correlation with system configuration parameters and operating condition parameters as independent variables; Step 4): Using the explicit correlation of the comprehensive performance as the performance evaluation kernel, and combining multi-objective optimization and multi-index objective decision-making methods, output the optimal system parameter configuration scheme that compromises between the performance indicators of each system.

[0009] In one embodiment, the artificial intelligence is to introduce the SHAP posterior interpretation method on the basis of the surrogate model, to perform additivity decomposition and dependency characterization of the contribution of input features to output performance indicators, and to construct a "model-interpretation" integrated method with explicit correlation of comprehensive performance.

[0010] In one embodiment, the compressed air energy storage system includes, but is not limited to, advanced adiabatic compressed air energy storage systems, afterburning compressed air energy storage systems, underwater compressed air energy storage systems, miniaturized / distributed compressed air energy storage systems, and liquid air energy storage. The system-level mechanism model includes a thermodynamic sub-model and an economic sub-model; wherein the thermodynamic sub-model includes at least a compression / expansion thermodynamic process model, a heat exchange and storage / regeneration loop model, and a gas storage-surrounding rock heat transfer model; the economic sub-model includes at least the calculation of equipment investment costs and operation and maintenance costs, the inclusion of energy purchase and auxiliary power costs, and discounting and annualization based on the discount rate and operating years. The design boundary and scenario parameters include one or more of the following: gas storage type and effective volume range, allowable operating pressure range, ambient and formation temperature, energy storage to release time ratio, charge / discharge power level and electricity price curve, project lifespan, and investment ceiling. The input parameters include at least: number of compression stages, number of expansion stages, initial pressure of the gas storage tank, ambient temperature, isentropic efficiency of the compressor, isentropic efficiency of the expander, external inlet mass flow rate, temperature of the medium-temperature heat transfer medium storage tank, temperature of the low-temperature heat transfer medium storage tank, radiator outlet air temperature, energy storage-to-release time ratio, and convective heat transfer coefficient of the gas storage tank wall. The system-level performance indicators include at least electro-electric efficiency, energy storage density, and cost per kilowatt-hour.

[0011] In one embodiment, the design boundaries and engineering constraints in step 2) include one or more of the following categories: operating condition parameters, including charging power, discharging power, service type, and allowable start-stop frequency or minimum start-stop time; gas storage and site parameters, including gas storage type and effective volume range, allowable working pressure range, and surrounding rock or stratum temperature field characteristics; and economic and policy parameters, including peak-valley or time-of-use electricity price curves, discount rate, project life, investment ceiling, and price or subsidy mechanisms related to energy storage projects.

[0012] In one embodiment, after obtaining the initial sample set in step 2), all input features are dimensionless to eliminate the influence of dimensional differences on surrogate model training and explicit correlation fitting.

[0013] In one embodiment, the AI ​​agent model in step 3) is constructed using a gradient boosting framework based on decision tree ensemble, preferably the Light Gradient Boosting Machine (LightGBM) model; its training process is conducted using a performance database sample set. The implementation is achieved by minimizing the squared loss function with regularization, whereby the loss function is: In the formula, S The total number of samples, K The number of system performance indicators. x (s) For the first s The system configuration parameters and operating condition parameter vectors for each sample y (s) This represents the true value of the corresponding system performance indicator. For the first k The true value of each system performance indicator For the proxy model to the first k Predicted values ​​of system performance indicators The regularization intensity coefficient is . The regularization term used to constrain model complexity is calculated as follows: In the formula, T This represents the number of decision trees in the gradient boosting model. L t For the first t The set of leaf nodes of a tree This represents the number of leaf nodes in the tree. w t,j For the first t Tree No. j The output weights of each leaf node and The regularization coefficient is used to balance the number of trees and the penalty strength of leaf node weights; through Bayesian optimization and early stopping strategies, adaptive search and cross-validation are performed on training configurations such as the number of trees, depth, number of leaf nodes, learning rate, and subsampling ratio, so that the obtained optimal surrogate model satisfies the coefficient of determination for predicting system performance indicators such as electricity-electric efficiency, energy storage density, and cost per kilowatt-hour on independent test sets. R 2 If the root mean square error is greater than 0.99, then the root mean square error is ( ). RMSE ) and mean absolute error ( MAE All three are significantly lower than the corresponding indicator units, thus accurately depicting the nonlinear mapping relationship between input parameters and overall system performance. The calculation methods for these three are as follows: In one embodiment, the process of explicitly expressing the model prediction relationship as an explicit correlation of comprehensive effectiveness in step 3) includes: First, applying the SHAP feature attribution method based on cooperative game theory to perform additive decomposition of the trained surrogate model based on single-value Shapley values, and identifying key configuration parameters and key operating condition parameters affecting the performance indicators of each system level through global importance ranking and SHAP dependency analysis. The calculation of Shapley values ​​is as follows: In the formula, ( x i This is the characteristic. x i The Shapley value reflects the average marginal contribution of this feature to the model's predictions. N The set of indices for all input features. N {i} Not including x i The set of all possible input features S for x i The feature subset in.

[0014] Based on this, the first k Predicted values ​​of system performance indicators This is expressed as the sum of the baseline value and the contributions of each input feature: In the formula, When all features do not contribute, the first k The model prediction baseline value of each system performance index For the first i The Shapley contribution of each input feature to the performance index. N The number of input features.

[0015] Secondly, according to the contribution function The dependency relationship will affect the first k The set of feature indices that exhibit an approximately linear influence on a performance index is defined as follows: L k The set of feature indices that exhibit monotonically nonlinear effects and diminishing marginal returns is defined as... N k At the same time, compression levels are introduced. n c With expansion series n e The coupling correction function is used to construct an explicit correlation of structured comprehensive effectiveness with clear physical meaning: In the formula, The input features have been dimensionless. These are fitting parameters used to scale the intensity of the influence of different nonlinear characteristics. These are the characteristic coefficients obtained through constrained nonlinear least squares fitting. To characterize the correction function of the coupling effect between the compression and expansion series, This is its internal parameter vector. The absolute value of the relative prediction deviation of the explicit correlation of the comprehensive performance within the feasible parameter domain is controlled to be within 10%.

[0016] In one embodiment, step 4), integrating the explicit correlation of comprehensive performance into a lightweight performance evaluation kernel and embedding it into a multi-objective optimization algorithm, means implementing the explicit correlation of comprehensive performance as a mathematical computation module that does not require calling the mechanistic model solver and does not rely on the interpretable artificial intelligence training process. In each fitness evaluation of the multi-objective optimization algorithm, the computation module is directly called to output performance indicators, reducing the computation time of a single scheme evaluation from minutes when calling the mechanistic model to seconds or milliseconds.

[0017] In one embodiment, step 4) of the multi-index objective decision-making method employs a combination of entropy weight-approximation ideal solution ranking (TOPSIS) to comprehensively evaluate and rank the optimized Pareto solution set: firstly, the entropy weight method is used to calculate the objective weights of each performance index. Let the Pareto solution set contain... M One candidate solution N One evaluation index, construction scheme – index matrix ,in x mn Indicates the first m The first scheme is in the n The values ​​for each indicator are calculated. After dimensionless processing of each indicator, the proportionality coefficient of each indicator is calculated: Based on this, we obtain the first... n The entropy value of each indicator: And further calculate the entropy weight of this index: Obtaining the entropy weights of each indicator w n Then, the TOPSIS method is used to calculate the relative closeness of each solution to the ideal solution. Specifically, the weighted decision matrix is ​​used to determine the... n The positive ideal solution of each indicator and negative ideal solution Calculate the first m The weighted Euclidean distances from each solution to the positive and negative ideal solutions are: And based on this, the first m The relative similarity of the two solutions: Ultimately based on proximity C m The magnitude of the parameter is used to sort the candidate solutions in the Pareto solution set, and the solution with the highest similarity is selected as the overall optimal parameter configuration solution, so as to achieve automatic selection of the optimal solution in terms of overall efficiency.

[0018] Compared with existing technologies, this invention constructs an integrated rapid design method for compressed air energy storage systems, encompassing "mechanism model – performance database construction – AI proxy model training and SHAP posterior interpretation – explicit correlation derivation of comprehensive performance – multi-objective optimization – objective decision-making based on multiple indicators." This method automates and clarifies the entire process from scenario requirements to parameter configuration. On one hand, by introducing an interpretable AI proxy model and explicit correlation of comprehensive performance, it maintains high-precision evaluation capabilities while endowing the model with clear physical meaning and parameter-performance analytical relationships, improving the engineering interpretability and traceability of optimization results. On the other hand, by combining a lightweight evaluation kernel with a multi-objective evolutionary algorithm, it significantly reduces computational overhead, enhances the depth and response speed of scheme exploration in the early engineering design stage, and, through an objectively weighted multi-indicator decision-making mechanism, achieves standardized, reproducible, and easily scalable comprehensive selection of compressed air energy storage system design schemes, providing an efficient and reliable technical approach for the planning and design of long-term energy storage projects. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall process of a rapid design method for compressed air energy storage system that integrates artificial intelligence and mechanistic modeling according to the present invention. Figure 2 This is a schematic diagram illustrating the specific implementation process of a rapid design method for compressed air energy storage systems that integrates artificial intelligence and mechanistic models, as described in this invention.

[0020] Figure 3 This diagram illustrates the closeness between the Pareto front for multi-objective optimization and the decision-making process obtained using the method of this invention. Detailed Implementation

[0021] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0022] like Figure 1 and Figure 2As shown, the rapid design method for the compressed air energy storage system of the present invention is as follows: First, for advanced adiabatic compressed air energy storage systems, key equipment such as multi-stage compressors, multi-stage expanders, heat exchangers, cooling towers, air storage tanks, and medium- and low-temperature heat medium storage tanks are modeled in units to establish a thermodynamic model-economic analysis coupled model for advanced adiabatic compressed air energy storage systems.

[0023] In this embodiment, the compression stage, expansion stage, ambient temperature, energy storage-release time ratio, compressor isentropic efficiency, expander isentropic efficiency, medium-temperature heat medium storage tank temperature, low-temperature heat medium storage tank temperature, initial pressure of gas storage tank, heat transfer efficiency of gas storage chamber wall, external air intake mass flow rate, and radiator outlet air temperature are used as input design parameters; and the electric-electric round-trip efficiency, energy storage density, and cost per kilowatt-hour are used as system-level comprehensive performance indicators, thus forming a multi-input-multi-output performance evaluation framework.

[0024] Secondly, the feasible parameter domain of the input parameters is determined by combining thermodynamic and heat transfer mechanism constraints with engineering and economic constraints. In this embodiment, the input parameter system and its upper and lower bounds are shown in Table 1. Within the feasible parameter domain, an input sample set is generated using the Latin hypercube sampling method, and the mechanism model is called for batch calculation to obtain a compressed air energy storage system performance database covering the feasible parameter domain.

[0025] Table 1. Parameter System and its Upper and Lower Bounds Next, a LightGBM model is trained based on the performance database. Hyperparameters such as the maximum tree depth, number of leaf nodes, and learning rate are adjusted through hyperparameter search and cross-validation to meet the requirements of prediction accuracy and stability for engineering applications. Furthermore, a cooperative game theory-based SHAP feature attribution method is employed to perform global and local interpretations of the trained LightGBM model, thereby obtaining the contribution ranking and influence direction of each input parameter to the performance index and identifying key influencing parameters. The SHAP values ​​of all input features are added to the prediction baseline values ​​of the LightGBM model to obtain the correlation formulas for the three core performance parameters. The final formula is as follows: surface As shown in Figure 2.

[0026] surface 2. System Comprehensive Performance Index Correlation Finally, the comprehensive performance is explicitly correlated and integrated into a lightweight performance evaluation kernel, embedded in the iterative process of a multi-objective optimization algorithm. With the goals of improving energy conversion efficiency and energy storage density, and reducing the levelized cost of electricity (LCOE), under the constraints of thermodynamic and heat transfer mechanisms as well as engineering and economic constraints, the NSGA-II algorithm is used to jointly optimize the system configuration parameters and operating condition parameters to obtain a multi-objective Pareto optimal solution set. Based on the objective weights of each performance index determined by the entropy weight method, the TOPSIS method is used to calculate the closeness. C m Based on this, the results are sorted in descending order, and the optimization results and the decision ranking of each scheme are as follows: Figure 3 As shown.

[0027] Compared with the original design of the power plant, the optimal compromise solution proposed by the method of this invention can improve the overall system efficiency by an average of about 17.37% while meeting engineering constraints, and the average error between the correlation calculation results and the numerical simulation results is only 2.01%. At the same time, the calculation time of the complete optimization design process is significantly reduced from about 12 hours for traditional numerical simulation methods and about 15 minutes for direct surrogate model-based methods to about 2 seconds, which is more than three orders of magnitude higher than the computational efficiency. This fully verifies that the method of this invention has significant efficiency and engineering applicability in the engineering design of compressed air energy storage systems.

[0028] This invention, based on the thermodynamic model-economic analysis coupled model of compressed air energy storage systems, introduces interpretable artificial intelligence methods to construct an integrated design framework consisting of "performance database-surrogate model training and feature attribution analysis-explicit correlational derivation of comprehensive performance-multi-objective optimization-objective decision-making with multiple indicators." It replaces time-consuming high-fidelity numerical simulations and "black box" surrogate models with structured explicit correlational methods, significantly reducing the computational cost of scheme optimization while maintaining high prediction accuracy and verifiable parameter contribution interpretation. This provides verifiable quantitative evidence for scheme comparison and parameter finalization in the early engineering design of compressed air energy storage systems.

Claims

1. A rapid design method for compressed air energy storage systems that integrates artificial intelligence and mechanistic models, characterized in that, Includes the following steps: Step 1): Construct a thermodynamic model-economic analysis coupled model of the compressed air energy storage system as a system-level mechanism model. Combine system configuration parameters and operating condition parameters to form input features, and use multiple system performance indicators that characterize energy conversion efficiency, energy storage space utilization rate and life cycle economy as output features. Step 2): Under the design boundary and engineering constraints, Latin hypercube sampling is used to generate sample points in the input parameter space, and the mechanism model is called to calculate the corresponding system-level performance index to construct a multi-dimensional performance database; Step 3): Construct a proxy model based on the performance database, and introduce the SHAP posterior interpretation method to decompose the feature contribution and characterize the dependency relationship of the proxy model, and express the model prediction relationship explicitly as a comprehensive performance explicit correlation with system configuration parameters and operating condition parameters as independent variables; Step 4): Using the explicit correlation of the comprehensive performance as the performance evaluation kernel, and combining multi-objective optimization and multi-index objective decision-making methods, output the optimal system parameter configuration scheme that compromises between the performance indicators of each system. The process of explicitly expressing the model prediction relationship as an explicit correlation of comprehensive effectiveness in step 3) is as follows: applying the SHAP method to perform single-value decomposition on the surrogate model output, and then... k Predicted values ​​of individual performance indicators This is expressed as the sum of the baseline value and the contributions of each input feature: In the formula, When all features do not contribute, the first k The model prediction baseline value of each system performance index For the first i The Shapley contribution of each input feature to the performance index. N The number of input features; Based on this, and according to the dependency relationship of contribution values, the first... k The set of feature indices that exhibit an approximately linear influence on a performance index is defined as follows: L k The set of feature indices that exhibit monotonically nonlinear effects and diminishing marginal returns is defined as... N k At the same time, compression levels are introduced. n c With expansion series n e The coupling correction function is used to construct an explicit correlation of comprehensive performance with clear physical meaning: In the formula, The input features have been dimensionless. These are fitting parameters used to scale the intensity of the influence of different nonlinear characteristics. These are the characteristic coefficients obtained through constrained nonlinear least squares fitting. To characterize the correction function of the coupling effect between the compression and expansion series, It is its internal parameter vector.

2. The rapid design method for compressed air energy storage systems integrating artificial intelligence and mechanistic models according to claim 1, characterized in that, The compressed air energy storage system described in step 1) includes an advanced adiabatic compressed air energy storage system, a supplementary combustion compressed air energy storage system, and an underwater compressed air energy storage system; the system-level mechanism model includes a thermodynamic sub-model and an economic sub-model; wherein, the thermodynamic sub-model includes a compression / expansion thermodynamic process model, a heat exchange and heat storage / regeneration loop model, and a gas storage-surrounding rock heat transfer model; the economic sub-model includes the calculation of equipment investment costs and operation and maintenance costs, the inclusion of energy purchase and auxiliary power costs, and the discounting and annualization based on the discount rate and operating years; the input parameters include one or more of the following parameters: number of compression stages, number of expansion stages, initial pressure of the gas storage tank, ambient temperature, compressor isentropic efficiency, expander isentropic efficiency, external inlet mass flow rate, medium-temperature heat medium storage tank temperature, low-temperature heat medium storage tank temperature, radiator outlet air temperature, energy storage-release time ratio, and gas storage tank wall convective heat transfer coefficient; the system-level performance indicators include electro-electric efficiency, energy storage density, and cost per kilowatt-hour.

3. The rapid design method for compressed air energy storage systems integrating artificial intelligence and mechanistic models according to claim 1, characterized in that, Its features are, The design boundaries and engineering constraints described in step 2) include one or more of the following categories: operating condition parameters, including charging power, discharging power, service type, and allowable start-stop frequency or minimum start-stop time; gas storage and site parameters, including gas storage type and effective volume range, allowable working pressure range, and surrounding rock or stratum temperature field characteristics; and economic and policy parameters, including peak-valley or time-of-use electricity price curves, discount rate, project life, investment ceiling, and price or subsidy mechanisms related to energy storage projects.

4. The rapid design method for compressed air energy storage systems integrating artificial intelligence and mechanistic models according to claim 1, characterized in that, Its features are, The AI ​​agent model described in step 3) is constructed using a gradient boosting framework based on decision tree ensemble, and the training process is performed on a performance database sample set. The implementation is achieved by minimizing the squared loss function with regularization, whereby the loss function is: In the formula, S The total number of samples, K The number of system performance indicators. x (s) For the first s The system configuration parameters and operating condition parameter vectors for each sample y (s) This represents the true value of the corresponding system performance indicator. For the first k The true value of each system performance indicator For the proxy model to the first k Predicted values ​​of system performance indicators The regularization intensity coefficient is . This is a regularization term used to constrain the complexity of the model.

5. The rapid design method for compressed air energy storage systems integrating artificial intelligence and mechanistic models according to claim 4, characterized in that, The formula for calculation is: In the formula, T This represents the number of decision trees in the gradient boosting model. L t For the first t The set of leaf nodes of a tree This represents the number of leaf nodes in the tree. w t,j For the first t Tree No. j The output weights of each leaf node and The regularization coefficient is used to balance the number of trees and the penalty strength of leaf node weights. The depth of the tree, the number of leaf nodes, the learning rate, and the subsampling ratio are adaptively searched through Bayesian optimization and early stopping strategies so that the obtained optimal surrogate model can meet the preset requirements for the prediction accuracy of the system performance index on the independent test set.

6. The rapid design method for compressed air energy storage systems integrating artificial intelligence and mechanistic models according to claim 4 or 5, characterized in that, The AI ​​agent model uses the LightGBM model.

7. The rapid design method for compressed air energy storage systems integrating artificial intelligence and mechanistic models according to claim 1, characterized in that, The absolute value of the relative deviation of the comprehensive performance explicit correlation within the feasible parameter domain is controlled to be within 10%.

8. The rapid design method for compressed air energy storage systems integrating artificial intelligence and mechanistic models according to claim 1, characterized in that, Step 4) describes using the explicit correlation formula of comprehensive performance as the performance evaluation kernel and embedding it into the multi-objective optimization algorithm. This means implementing the explicit correlation formula of comprehensive performance as a mathematical calculation module that does not require calling the mechanism model solver and does not depend on the interpretable artificial intelligence training process. In each fitness evaluation of the multi-objective optimization algorithm, the calculation module is directly called to output the performance index, thereby reducing the calculation time of a single scheme evaluation from minutes when calling the mechanism model to seconds or milliseconds.

9. The rapid design method for compressed air energy storage systems integrating artificial intelligence and mechanistic models according to claim 1, characterized in that, The multi-index objective decision-making method described in step 4) adopts a combination of entropy weight-approximation ideal solution ranking (TOPSIS) to comprehensively evaluate and rank the Pareto solution set obtained by optimization: the objective weight of each performance index is determined by the entropy weight method, and the relative closeness of each scheme to the ideal solution is calculated by combining the TOPSIS method. The scheme with the largest closeness is selected as the comprehensive optimal parameter configuration scheme.

Images