System and method for dynamic utilization of cold energy of gas storage reservoir in million cubic meter compressed air energy storage system

Abstract

The present invention provides a system and method for dynamically utilizing cold energy in a million-cubic-meter compressed air energy storage system. The method comprises: acquiring multi-source monitoring data to construct a basic data set; applying a multi-scale thermal flow constraint space reconstruction method to establish a spatial-temporal dynamic model of the gas storage reservoir; developing a hierarchical control framework to achieve coordinated control of the throttle valve and the cold energy recovery device; and constructing a method for enhancing operating condition similarity based on cold energy characteristics and a method for adaptively generating a cold energy recovery control strategy. This invention solves technical problems such as accurately modeling the temperature field within large gas storage reservoirs, identifying cold energy characteristic operating conditions, and coordinating throttling and cold energy recovery control. It improves cold energy recovery efficiency by 20-30% and reduces turbine inlet temperature fluctuations.

Description

Technical Field

[0001] The present invention relates to CAES technology, and in particular to a system and method for dynamically utilizing cold energy from a gas storage reservoir of a million-cubic-meter compressed air energy storage system. Background Art

[0002] As the proportion of renewable energy connected to the grid continues to increase, the power system is increasingly in need of large-scale energy storage technologies. Compressed air energy storage (CAES), with its advantages of large capacity, long life, and low cost, has become a key technology for large-scale energy storage. In a million-cubic-meter CAES system, the degassing process of the gas storage generates a large amount of low-temperature cold energy. Effectively recovering and utilizing this cold energy not only improves the system's overall energy utilization efficiency but also reduces thermal shock in the first-stage heat exchanger of the turbine stage, enhancing the overall safety and economic efficiency of the system.

[0003] Current research on cold energy utilization in compressed air energy storage systems is mainly focused on small systems, using conventional throttling control and simple cold energy recovery methods. Traditional methods usually use a throttle valve positioning system based on PID control, which adjusts the throttle valve opening by measuring the turbine inlet temperature and pressure to ensure stable operation of the turbine. In terms of cold energy recovery, common technical means include direct use of cold air for cooling, coupling with heat pump systems, and application modes such as industrial cooling sources. In terms of spatial modeling, temperature field calculations based on finite element or finite volume methods are mainly used, while the control strategy is mainly based on rule-based control algorithms. Operating condition identification often uses feature comparison methods under a single time scale.

[0004] However, when cold energy recovery devices are incorporated into million-cubic-meter CAES systems, traditional approaches face numerous technical challenges. For example, the complex coupling between the throttle valve and the cold energy recovery device makes it difficult for traditional independent control strategies to balance the dual objectives of stabilizing turbine inlet parameters and maximizing cold energy recovery efficiency. This strategy performs particularly poorly in dynamic processes with rapidly changing system states, severely hindering both improved cold energy recovery efficiency and stable operation. Summary of the Invention

[0005] Purpose of the invention: To provide a system and method for dynamically utilizing cold energy of a million cubic meter compressed air energy storage system gas storage reservoir, in order to solve at least one technical problem existing in the prior art.

[0006] Technical solution: Dynamic utilization method of cold energy in a million cubic meter compressed air energy storage system, including:

[0007] Acquire multi-source monitoring data of the compressed air energy storage system to form a basic data set for the system; based on this, reconstruct the multi-scale thermal flow constraint space and generate a dynamic cooling energy distribution prediction map through a space-time dynamic model;

[0008] Based on the basic data set of the system, a hierarchical control framework is constructed through the coupling impact model of throttling and cold energy recovery to generate a hierarchical control strategy;

[0009] Based on the system basic data set and dynamic cooling energy distribution prediction map, a working condition feature fingerprint library is constructed, and a cooling energy recovery optimization control strategy is generated by identifying similar working conditions; it is integrated with the hierarchical control strategy to output a comprehensive control instruction set.

[0010] According to one aspect of the present application, generating a dynamic cooling energy distribution prediction map includes:

[0011] Based on the system basic data set, the full-space temperature and pressure fields in the gas storage are preliminarily reconstructed. Based on this, multi-scale thermal flow constraint space reconstruction is performed to optimize the full-space temperature and pressure fields.

[0012] Based on the optimized full-space temperature and pressure fields, time-space coupled evolution prediction is performed to generate a dynamic cold energy distribution prediction map.

[0013] According to one aspect of the present application, optimizing the temperature field and pressure field of the entire space includes:

[0014] The preliminary reconstruction results of the full-space temperature field and pressure field are subjected to hierarchical spatial segmentation to obtain a multi-scale spatial domain; based on this, physical constraint models for different scales are constructed to form a multi-scale physical constraint set;

[0015] The gas storage wall temperature sensor data in the system basic data set is used to construct a dynamic boundary condition set through adaptive boundary constraints. This is combined with the sensor measured data and multi-scale physical constraint set in the system basic data set to perform physics-guided interpolation reconstruction and optimize the full-space temperature and pressure fields.

[0016] According to one aspect of the present application, a time-space coupled evolution prediction is performed to generate a dynamic cold energy distribution prediction map, including:

[0017] Based on the optimized full-space temperature and pressure fields, adaptive time window analysis is applied to identify the temporal dynamic characteristics of different stages and generate a set of temporal characteristic segments. For each stage, a temperature gradient-sensitive spatial evolution model is constructed to form a set of segmented spatial evolution models.

[0018] The bleed parameters of the throttle valve operating state sequence in the system basic data set are read, the impact of throttle valve opening changes on the evolution of the temperature field inside the gas storage reservoir is analyzed, and a bleed-temperature coupling model is established. This model is combined with the segmented spatial evolution model set to generate a dynamic cold energy distribution prediction map through multi-step progressive prediction.

[0019] According to one aspect of the present application, generating a temporal characteristic segment set includes:

[0020] Based on the optimized full-space temperature and pressure fields, the statistical characteristics of the full-space temperature change rate, including mean, variance, and kurtosis, are calculated.

[0021] Through time series clustering, statistical features are classified to automatically identify rapid change, transition and slow change stages; the time step is adjusted according to the characteristics of different stages: the first small time step is used to improve prediction accuracy in the rapid change stage, and the second time step is used to improve calculation efficiency in the slow change stage; finally, a set of time characteristic segments is generated, in which the first time step is smaller than the second time step.

[0022] According to one aspect of the present application, generating a cold energy recovery optimization control strategy includes:

[0023] Based on the basic system data set, a working condition feature fingerprint library of the compressed air energy storage system is constructed; based on this, a working condition similarity vector is generated from the current system state through working condition identification;

[0024] Combined with the dynamic cooling energy distribution prediction map, the operating condition similarity vector is enhanced based on the cooling energy characteristics to obtain the optimized operating condition similarity vector; based on this, the cooling energy recovery control strategy is adaptively generated and the cooling energy recovery optimization control strategy is output.

[0025] According to one aspect of the present application, obtaining an optimized operating condition similarity vector includes:

[0026] Based on the system basic data set and dynamic cooling energy distribution prediction map, the cooling energy characteristic indicators are extracted, the similarity calculation formula for cooling energy recovery is constructed, and a thermodynamic similarity index set is generated;

[0027] According to the characteristics of the cooling energy recovery process, a stage-by-stage similarity weighting strategy is constructed to generate a stage-by-stage weighting function set.

[0028] Read the cold energy recovery device operating data in the system basic data set, analyze the relationship between operating condition similarity and cold energy recovery efficiency, and establish an efficiency correlation map;

[0029] Combining the thermodynamic similarity index set, stage weighting function set and efficiency association mapping, the enhanced similarity between the current and historical operating condition similarity vectors is calculated, and the optimized operating condition similarity vector is output.

[0030] According to one aspect of the present application, the set of thermodynamic similarity indicators includes:

[0031] The temperature gradient distribution similarity index is obtained by calculating the cosine similarity of the temperature gradient fields of two working conditions;

[0032] The similarity index of the motion pattern of the hot and cold interfaces is obtained by comparing the movement speed and direction of the hot and cold interfaces;

[0033] The similarity index of the cold energy generation rate is obtained by comparing the change pattern of the cold energy generation per unit time.

[0034] According to one aspect of the present application, a cold energy recovery control strategy is adaptively generated and a cold energy recovery optimization control strategy is output, including:

[0035] Based on the optimized operating condition similarity vector, control strategies for similar operating conditions are retrieved from the historical control database, key control decisions are extracted, and a set of candidate control strategies is generated. From these strategies, control decision factors related to cooling energy recovery efficiency are identified, and a set of cooling energy efficiency influencing factors is constructed.

[0036] Based on the coupling influence model and the set of cooling energy efficiency influencing factors, a throttling-recovery dual-objective optimization framework is constructed. The throttling valve control and the cooling energy recovery device control are regarded as coupled optimization problems, and a multi-objective optimization model is constructed. It is combined with the current system status to perform real-time optimization calculations and generate the cooling energy recovery optimization control strategy under the current operating conditions.

[0037] The million cubic meter compressed air energy storage system gas storage cold energy dynamic utilization system includes:

[0038] Compression system, used to compress air;

[0039] Gas storage, connected to the compression system, used to store compressed gas;

[0040] An exhaust pipe connected to the gas storage reservoir and used to discharge the gas in the gas storage reservoir;

[0041] A throttle valve is provided in the exhaust pipe to control the gas flow and pressure;

[0042] A cold energy recovery device is provided at the rear end of the exhaust port of the throttle valve in the exhaust pipe and is used to recover cold energy in the gas;

[0043] The heat exchanger is installed at the rear end of the cold energy recovery device and exchanges heat with the gas through cold energy;

[0044] The turbine system is installed after the heat exchanger and is used to generate electricity by turning the gas after heat exchange;

[0045] Control module, including:

[0046] at least one processor; and,

[0047] a memory communicatively connected to at least one of the processors; wherein,

[0048] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the method of the above technical solution.

[0049] Beneficial effects: The present invention ensures that the cold energy recovery device can be effectively arranged and controlled according to the actual cold energy distribution position, avoiding energy waste and inefficient equipment operation; improves the cold energy recovery efficiency and system operation stability; ensures the stable operation of the turbine system, resolves the contradiction between cold energy recovery and stable turbine operation, and improves the overall energy utilization efficiency of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of the steps of a method for dynamically utilizing cold energy from a gas storage reservoir in a million-cubic-meter compressed air energy storage system provided in an embodiment of the present application.

[0051] Figure 2 A flowchart of the steps for optimizing the temperature and pressure fields in the entire space provided in an embodiment of the present application.

[0052] Figure 3 A flowchart of the steps for obtaining an optimized operating condition similarity vector provided in an embodiment of the present application.

[0053] Figure 4 A flowchart of the steps for adaptively generating an output cooling energy recovery optimization control strategy by performing cooling energy recovery control strategy according to an embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0055] It should be noted that to clearly illustrate the steps of this application, serial numbers are assigned to each step in the specification. These serial numbers are for illustrative purposes only and do not limit the order in which the steps must be executed. In actual operation, depending on the technical requirements of the specific implementation scenario, the steps may be executed in a different order than shown in the specification, and in some cases, parallel processing between steps may be implemented.

[0056] During the research process, it was found that there is obvious temperature stratification and uneven distribution inside large gas storage facilities. Traditional spatial modeling methods cannot accurately capture such complex temperature fields under sparse sampling point conditions, resulting in inaccurate predictions of cold energy locations. In addition, conventional operating condition identification methods lack specific considerations for cold energy characteristics, making it difficult to identify operating conditions with similar cold energy recovery efficiency, affecting the effective transfer of control experience.

[0057] Specifically, in the compressed air energy storage systems currently under construction and research, the cooler air released from the gas storage is directly heated before entering the turbine for turbine operation, without utilizing the cold energy of the exhaust gas from the gas storage. For small-scale compressed air energy storage power plants, the recovery of cold energy from the gas storage is relatively low in terms of both economic and practical value. However, for million-cubic-meter-scale compressed air energy storage power plants, it has both economic and practical value. To fully utilize the cold energy of the gas storage in compressed air energy storage power plants and improve the overall energy utilization efficiency of the system, a dynamic cold energy utilization system for the gas storage of million-cubic-meter-scale compressed air energy storage systems has been proposed.

[0058] The compressed air energy storage power station primarily consists of a gas storage reservoir, an exhaust pipe, a throttle valve, a cold energy recovery device, a heat exchanger, a turbine system, and a compression system. During power generation, high-pressure gas is discharged through the exhaust pipe. The discharged high-pressure air then passes through the throttle valve and is injected into the cold energy recovery device. The high-pressure air then passes through the cold energy recovery system and is injected into the heat exchanger before entering the turbine system to generate electricity. The cold energy recovery device is connected to the exhaust port of the throttle valve in the gas storage reservoir's exhaust pipe. It uses a medium to recover cold energy from the exhaust air within the gas storage reservoir and stores it in a cold energy storage tank. The amount of recovered cold energy can be controlled by adjusting the throttle valve and the gas storage reservoir's exhaust flow rate. Specifically, the compression system is used to compress the air; the gas storage is connected to the compression system for storing the compressed gas; the exhaust pipe is connected to the gas storage for discharging the gas in the gas storage; the throttle valve is arranged in the exhaust pipe for controlling the gas flow and pressure; the cold energy recovery device is arranged at the rear end of the exhaust port of the throttle valve in the exhaust pipe for recovering the cold energy in the gas; the heat exchanger is arranged at the rear end of the cold energy recovery device for exchanging heat with the gas through the cold energy; the turbine system is arranged after the heat exchanger for generating electricity by turbine operation on the gas after heat exchange.

[0059] like Figure 1 As shown, for the above system, a method for dynamically utilizing cold energy of a gas storage reservoir of a million cubic meter compressed air energy storage system is provided, comprising:

[0060] S1. Acquire multi-source monitoring data of the compressed air energy storage system to form a basic data set of the system;

[0061] Specifically, the multi-source monitoring data includes multi-point temperature and pressure data inside the gas storage reservoir, throttle valve operating status data, and cold energy recovery device operating parameters.

[0062] S2. Based on the system basic data set, multi-scale heat flow constraint space reconstruction is performed, and a dynamic cold energy distribution prediction map is generated through a space-time dynamic model;

[0063] Specifically, multi-scale thermal flow constraint spatial reconstruction means the system considers data at different levels, such as microscopic temperature variations and macroscopic overall energy trends, to analyze and optimize spatial thermal flow and more accurately simulate temperature and energy flows within the compressed air energy storage system. The space-time dynamic model combines spatial (regions within the storage system) and temporal (different operating phases) variations to dynamically predict the flow and distribution of cold energy.

[0064] S3. Based on the basic data set of the system, a hierarchical control framework is constructed through the coupling impact model of throttling and cold energy recovery to generate a hierarchical control strategy;

[0065] Specifically, throttling refers to the restriction of air flow through the energy storage system, which affects pressure and temperature changes. Cold energy recovery refers to the recovery of low-temperature energy from the system to improve overall energy utilization. The two interact in a complex way, so mathematical models are needed to analyze how they influence each other and ensure that the system maintains efficient operation under different operating conditions. Because the system's operation involves multiple variables, such as pressure, temperature, and flow, direct control is complex, so a hierarchical control architecture is required. This framework may include multiple levels, such as: low-level control: regulating the operation of individual devices, such as the opening and closing status of throttle valves; mid-level control: comprehensively regulating the interactions between multiple devices, such as the coordinated operation of gas storage and cold energy recovery devices; and high-level control: performing overall optimization to ensure that the entire system meets the desired energy efficiency targets.

[0066] S4. Based on the system basic data set and dynamic cooling energy distribution prediction map, a working condition feature fingerprint library is constructed, and a cooling energy recovery optimization control strategy is generated by identifying similar working conditions; the cooling energy recovery optimization control strategy is integrated with the hierarchical control strategy to output a comprehensive control instruction set, including the throttle valve control signal and the cooling energy recovery device control signal.

[0067] Specifically, the operating condition fingerprint library is a database of different operating states of the system. Through this fingerprint library, the system can identify whether the current operating condition is similar to certain known states in the past. This allows the system to predict the current cooling energy flow based on historical experience and generate an optimized control strategy.

[0068] According to one aspect of the present application, the step of forming a system basic data set includes:

[0069] S11. Collect temperature and pressure data at multiple points inside the gas storage reservoir to construct a spatially distributed monitoring data set;

[0070] S12, acquiring throttle valve operating status data to form a throttle control data set;

[0071] S13, monitoring the operating parameters of the cold energy recovery device and establishing a cold energy recovery characteristic data set;

[0072] S14: Integrate the spatial distribution monitoring dataset, the throttling control dataset, and the cold energy recovery characteristic dataset, perform time alignment and cleaning, and output a system comprehensive monitoring dataset as the system basic dataset.

[0073] Specifically, temperature and pressure data from multiple points within the gas storage are collected to construct a spatially distributed monitoring dataset. Data from the distributed temperature sensor array (T1, T2, ..., Tn) and pressure sensor array (P1, P2, ..., Pm) within the gas storage are read and combined with the sensor spatial location information (X1, Y1, Z1, ..., Xn, Yn, Zn) to generate raw spatially distributed data with timestamps. Throttle valve operating status data is acquired to form a throttle control dataset. Throttle valve opening signal (Vo), pressure differential across the throttle valve (ΔP), temperature differential across the throttle valve (ΔT), and flow rate data (Q) are collected and combined with timestamp information to generate a throttle valve operating status sequence. The operating parameters of the cold energy recovery device are monitored to construct a cold energy recovery characteristic dataset. Key parameters such as the cold energy recovery medium inlet temperature (Tci), outlet temperature (Tco), medium flow rate (Qc), and heat exchange efficiency (η) of the recovery device are collected to generate cold energy recovery device operating data. Multi-source data is integrated, time-aligned, and cleaned to output a standardized system-wide monitoring dataset D. A timestamp alignment algorithm is applied to handle the sampling frequency differences between different data sources to ensure the temporal consistency and availability of data.

[0074] According to one aspect of the present application, the step of generating a dynamic cooling energy distribution prediction map includes:

[0075] S21. Preliminary reconstruction of the full-space temperature and pressure fields within the gas storage based on the system basic data set;

[0076] S22. Based on the preliminary reconstruction results, we conducted multi-scale thermal flow constraint space reconstruction to address the sparse monitoring point problem in million-cubic-meter gas storage facilities and optimized the temperature and pressure fields in the entire space.

[0077] S23. Based on the optimized full-space temperature field and pressure field, a time-space coupled evolution prediction is performed to generate a dynamic cold energy distribution prediction map.

[0078] Specifically, based on the system's comprehensive monitoring dataset D, a kernel interpolation method was applied to preliminarily reconstruct the full-space temperature field Ts(x, y, z, t) and pressure field Ps(x, y, z, t) within the gas storage. A basic spatial distribution model was constructed using standard kernel interpolation to provide initial values for subsequent optimization. To address the sparse monitoring points in the million-cubic-meter gas storage, a multi-scale thermal flow constraint spatial reconstruction method was developed to optimize the full-space temperature field Ts and pressure field Ps. Specifically, the method introduces constraints from fluid thermodynamics theory and uses predictions based on physical laws instead of pure mathematical interpolation in data-sparse areas, addressing the "false distribution" problem caused by insufficient sampling points in large gas storage reservoirs caused by traditional spatial reconstruction methods. A multi-scale analysis framework is constructed, applying different physical models at different spatial scales: a heat diffusion model is used to describe local temperature changes at the microscale, a convection heat transfer model is used to describe heat transfer caused by airflow at the mesoscale, and a laminar flow model is used to describe the overall temperature stratification phenomenon at the macroscale. An adaptive boundary constraint mechanism is constructed, using data from gas storage wall temperature sensors to construct dynamic boundary conditions, addressing the problem of unclear boundary conditions in traditional methods when dealing with large enclosed spaces.

[0079] A time-series-space coupled evolution prediction algorithm is constructed to generate a dynamic cold energy distribution prediction map G within the gas storage reservoir. This algorithm addresses the unique needs of cold energy recovery scenarios and overcomes the shortcomings of traditional spatiotemporal models: a non-uniform time step prediction mechanism is constructed to adaptively adjust the prediction time step based on the dynamic characteristics of different stages in the cold energy recovery process. Small time steps are used to improve accuracy in rapidly changing stages (such as the initial stage of deflation), while large time steps are used to improve efficiency in slowly changing stages. A temperature gradient-sensitive spatial evolution model is constructed, particularly enhancing prediction accuracy in low-temperature areas and areas with large temperature gradients, providing precise spatial positioning support for cold energy recovery. A coupled analysis of deflation flow and temperature field is introduced to establish a model for the impact of throttle valve opening changes on the evolution of the temperature field within the gas storage reservoir, accurately predicting changes in cold energy distribution under different deflation strategies.

[0080] like Figure 2 As shown, according to one aspect of the present application, the steps of optimizing the temperature field and pressure field of the entire space include:

[0081] The preliminary reconstruction results of the full-space temperature field and pressure field are subjected to hierarchical spatial segmentation to obtain a multi-scale spatial domain;

[0082] Based on the multi-scale spatial domain, physical constraint models for different scales are constructed to form a multi-scale physical constraint set;

[0083] Using the gas storage reservoir wall temperature sensor data in the system basic data set, a dynamic boundary condition set is constructed through adaptive boundary constraints;

[0084] The dynamic boundary condition set is combined with the sensor measured data and multi-scale physical constraint set in the system basic data set to perform physics-guided interpolation reconstruction to optimize the temperature and pressure fields in the entire space.

[0085] According to one aspect of the present application, the steps of obtaining a multi-scale spatial domain and forming a multi-scale physical constraint set include:

[0086] The million-cubic-meter gas storage space is divided into three scale levels: micro, meso and macro.

[0087] Create an adaptive grid for each layer, forming a fine grid in areas with large temperature gradients and a coarse grid in areas with gentle temperature changes;

[0088] At the microscale, the heat diffusion equation is used to describe local heat conduction. At the mesoscale, the convection heat transfer model is used to characterize the heat transfer caused by airflow. At the macroscale, the laminar flow model is used to capture the vertical temperature stratification phenomenon.

[0089] Boundary matching conditions are set for each physical model to ensure the consistency of models of different scales in overlapping areas and form a multi-scale physical constraint set.

[0090] Specifically, the preliminary reconstruction results of the full-space temperature field Ts and pressure field Ps are read, and the hierarchical space segmentation algorithm is applied to generate the multi-scale spatial domain partition D_scale. First, the million-cubic-meter gas storage space is divided into microscopic (0.1-1m 3 ), Middle View (1-100m 3 ) and macroscopic (>100m 3 ) three scale levels, and an octree partitioning method is used to create an adaptive grid for each level, forming a fine grid in areas with large temperature gradients and a coarse grid in areas with gentle temperature changes, to ensure efficient allocation of computing resources. Based on the multi-scale spatial domain partition D_scale, a physical constraint model for different scales is constructed to form a multi-scale physical constraint set P_constraint. At the microscale, the heat diffusion equation (ΨT / Ψt = α ▽ 2 T) describes the local heat conduction; at the mesoscale, the convection heat transfer model (ΨT / Ψt + u· ▽ T = α ▽ 2 T) characterizes heat transfer caused by airflow; at the macroscale, a laminar flow model is used to capture vertical temperature stratification. Boundary matching conditions are set for each physical model to ensure consistency in overlapping areas of different scale models.

[0091] Combining the sensor measured data in the system comprehensive monitoring data set D with the multi-scale physical constraint set P_constraint, a physical-guided interpolation reconstruction is performed to optimize the full-space temperature field Ts and pressure field Ps. Unlike the traditional purely data-driven interpolation method, physical constraints are introduced in the interpolation process. The specific implementation is as follows: first, a finite element grid with sensor locations as nodes is constructed, and the physical constraint equations are discretized into a set of linear equations. Then, the measured data are used as boundary conditions, and the constrained least squares problem (min‖Ax-b‖ 2 +λ‖Lx‖ 2 ) to obtain the interpolation result that satisfies the physical constraints, where A is the observation matrix, L is the physical constraint operator, and λ is the balance parameter.

[0092] An adaptive boundary constraint mechanism is constructed, and the dynamic boundary condition set B_condition is constructed using the gas storage wall temperature sensor data in the original spatial distribution data. To address the difficulty in accurately determining the boundary conditions of million-cubic-meter gas storage reservoirs, a boundary learning algorithm based on measured data is constructed: first, the time-varying temperature distribution characteristics are extracted from the wall sensor data. Then, the heat conduction theory is used to deduce the heat exchange relationship between the wall and the internal fluid, and a non-uniform Neumann boundary condition that varies with time is constructed. Finally, this boundary condition is applied to the reconstruction process, solving the problem of inaccurate boundary condition setting in traditional methods when dealing with large enclosed spaces. This embodiment improves the accuracy of the reconstruction results from 85% of traditional methods to over 95%, especially in areas close to the wall, where the improvement is more significant.

[0093] According to one aspect of the present application, the steps of performing time-space coupled evolution prediction and generating a dynamic cold energy distribution prediction map include:

[0094] Based on the optimized full-space temperature and pressure fields, adaptive time window analysis is applied to identify the temporal dynamic characteristics of different stages and generate a temporal characteristic segmentation set.

[0095] For each stage in the temporal characteristic segmentation set, a temperature gradient-sensitive spatial evolution model is constructed to form a segmented spatial evolution model set;

[0096] Read the deflation parameters of the throttle valve operating state sequence in the system basic data set, analyze the impact of throttle valve opening changes on the evolution of the temperature field inside the gas storage, and establish a deflation-temperature coupling model;

[0097] The outgassing-temperature coupling model is combined with the segmented spatial evolution model set to generate a dynamic cold energy distribution prediction map through multi-step progressive prediction.

[0098] According to one aspect of the present application, the step of generating a temporal characteristic segment set includes:

[0099] Based on the optimized full-space temperature and pressure fields, the statistical characteristics of the full-space temperature change rate, including mean, variance, and kurtosis, are calculated.

[0100] Classify statistical features through time series clustering to automatically identify rapid changes, transitions, and slow changes;

[0101] The time step is adjusted according to the characteristics of different stages: a small time step within a preset range is used to improve prediction accuracy in the fast-changing stage, and a large time step within a preset range is used to improve computational efficiency in the slow-changing stage; ultimately, a temporal characteristic segment set is generated.

[0102] Based on the current temperature field, the spatial temperature gradient distribution is calculated, high gradient areas are identified, and an adaptive weighted partial differential equation solver is constructed, using finer spatial discretization and more complex flow models for high gradient areas.

[0103] Specifically, based on the optimized full-space temperature field Ts and pressure field Ps, adaptive time window analysis is applied to identify the temporal dynamic characteristics of different stages and generate a set of temporal characteristic segments T_segments. First, the statistical characteristics (mean, variance, kurtosis, etc.) of the full-space temperature change rate are calculated. Then, through the time series clustering of the change rate characteristics, the rapid change stage (such as the initial stage of deflation), the transition stage and the slow change stage (such as the end of deflation) are automatically identified. An appropriate time step is assigned to each stage, which not only ensures the prediction accuracy of the rapid change stage, but also improves the computational efficiency of the slow change stage. For each stage in the temporal characteristic segment set T_segments, a spatial evolution model sensitive to temperature gradients is developed to form a set of segmented spatial evolution models E_models. Unlike traditional spatiotemporal prediction models that treat all spatial regions equally, this embodiment particularly strengthens the evolution prediction of low-temperature areas and areas with large temperature gradients. The specific implementation is as follows: first, the spatial temperature gradient distribution is calculated based on the current temperature field to identify high-gradient areas; then, an adaptive weighted partial differential equation solver is constructed to adopt finer spatial discretization and more complex flow models for high-gradient areas; finally, for low-temperature areas, special consideration is given to the influence of phase change effects and condensation phenomena on temperature evolution to ensure the prediction accuracy of cold energy generation areas.

[0104] Read the bleed parameters in the throttle valve operating state sequence, analyze the impact of the throttle valve opening change on the evolution of the temperature field inside the gas storage, and establish a bleed-temperature coupling model C_model. Treat the throttling process and the evolution of the temperature field in the reservoir as a coupled system: first, construct a throttle valve mathematical model, and calculate the bleed flow rate and temperature drop based on the throttle valve opening and the pressure difference before and after the throttle valve; then track the flow field and temperature field changes around the bleed port, and establish a quantitative relationship between the bleed flow rate and the local temperature drop; finally, integrate this relationship into the overall temperature field evolution equation to accurately capture the changes in cold energy distribution under different bleed strategies. This embodiment solves the problem that traditional methods ignore the impact of the bleed dynamic process on the internal temperature field, and provides a more accurate prediction basis for cold energy recovery.

[0105] By integrating the segmented spatial evolution model set E_models and the degassing-temperature coupling model C_model, a multi-step progressive prediction algorithm is employed to generate a prediction map G of the dynamic cold energy distribution within the gas storage reservoir. First, based on the current temperature field and the preset degassing strategy, a short-term prediction is performed using the most appropriate time step. The prediction results are then used as new initial conditions and the prediction process is iteratively executed. Finally, by identifying areas with temperatures below the ambient temperature and their temperature differences, the cold energy density and total amount of each spatial unit are calculated, forming a spatiotemporally resolved cold energy distribution prediction map. Compared with traditional prediction methods, this embodiment reduces the prediction error from 15-20% to 5-8% when predicting the cold energy distribution of a million-cubic-meter gas storage reservoir. In particular, the accuracy of prediction for the location of concentrated cold energy areas is improved by over 40%, providing critical support for the precise control of the cold energy recovery device.

[0106] According to one aspect of the present application, the step of generating a hierarchical control strategy includes:

[0107] S31. Based on the throttling valve operating state sequence and the cold energy recovery device operating data in the system basic data set, establish a coupling influence model of throttling and cold energy recovery;

[0108] S32. Design a hierarchical control framework and generate a hierarchical control strategy based on the coupling impact model;

[0109] S33. Based on the hierarchical control strategy, a collaborative control algorithm is developed to output throttle valve optimization control signals and working fluid flow control signals to form a comprehensive control instruction set.

[0110] Specifically, a coupled influence model C of throttling and cold energy recovery is established based on the throttle valve operating state sequence and cold energy recovery device operating data. The dynamic impact of throttle valve parameter changes on the performance of the cold energy recovery device is analyzed, providing a theoretical basis for collaborative control. Based on the coupled influence model C, a hierarchical control framework is constructed, and a hierarchical control strategy P is generated. The control task is decomposed into control layers of different time scales, and control objectives are rationally allocated based on the different dynamic characteristics of the throttle valve and the cold energy recovery device. A collaborative control algorithm is constructed to output the optimized throttle valve control signal Vo and the working fluid flow control signal Qc. The algorithm comprehensively considers the turbine operating requirements and cold energy recovery efficiency to achieve a balanced optimization of the two.

[0111] The specific process for constructing the coupled impact model of throttling and cold energy recovery involves reading the throttling valve operating state sequence and cold energy recovery device operating data, applying time alignment and data preprocessing techniques to generate the coupled analysis dataset C_data. First, time normalization is performed to ensure temporal consistency across different data sources. Then, a sliding window method is used to segment the time series into multiple operating condition segments. Finally, each operating condition segment is normalized and detected for outliers to provide a high-quality data foundation for subsequent coupled analysis. Based on the coupled analysis dataset C_data, Granger causality analysis is used to identify the causal relationship between throttling valve parameters and cold energy recovery performance, and a dynamic causal network D_causal is constructed. Unlike traditional correlation analysis, this embodiment uses time lag analysis to determine the causal direction between variables. First, a multivariate time series model is constructed that includes factors such as throttle valve opening, front-to-rear pressure differential, and flow rate, as well as indicators such as cold energy recovery device inlet temperature, working fluid temperature difference, and recovery efficiency. The Granger causality index is then calculated by comparing the prediction errors of the restricted model with the complete model. Finally, a directed weighted network is constructed based on the causal index to reveal the dynamic causal relationship between throttling control and cold energy recovery. Based on the dynamic causal network D_causal, a coupling influence model C of throttling and cold energy recovery is constructed. The identified key causal paths are quantified into mathematical models: first, for each major causal path, the transfer function from the input variable (such as the change in throttle valve opening) to the output variable (such as the change in cold energy recovery efficiency) is extracted; then, multiple paths are integrated into a state-space model to describe the dynamic characteristics of the entire system; finally, historical data verification and parameter optimization are used to ensure the accuracy and generalization ability of the model. This embodiment breaks through the limitations of traditional independent control and provides a theoretical basis for the coordinated control of the throttle valve and the cold energy recovery device, improving the cold energy recovery efficiency by 15-25%.

[0112] According to one aspect of the present application, the step of generating a cold energy recovery optimization control strategy includes:

[0113] S41. Based on the system basic data set, build a working condition feature fingerprint library of the compressed air energy storage system;

[0114] S42. Based on the working condition feature fingerprint library, generate a working condition similarity vector from the current system state through working condition identification;

[0115] S43. In combination with the dynamic cooling energy distribution prediction map, the operating condition similarity vector is enhanced based on the cooling energy characteristics to obtain an optimized operating condition similarity vector;

[0116] S44. Based on the optimized operating condition similarity vector, a cooling energy recovery control strategy is adaptively generated, and the cooling energy recovery optimization control strategy is output.

[0117] Specifically, based on the system's comprehensive monitoring dataset D, a compressed air energy storage system operating condition fingerprint library H was constructed. Feature extraction and cluster analysis methods were applied to identify typical operating condition categories from historical operating data. A working condition identification algorithm was developed to generate a working condition similarity vector S based on the current system state. The similarity between the current working condition and each typical working condition in the working condition fingerprint library H was calculated to provide a basis for subsequent control strategy generation. A working condition similarity enhancement method based on cooling energy characteristics was developed to improve the targeted identification of similar working conditions. The specific implementation is as follows: In response to the shortcomings of traditional similar working condition identification in cold energy recovery applications, a thermodynamic similarity index system is constructed, and feature dimensions centered on cold energy generation and transfer are introduced into the traditional working condition similarity calculation, such as temperature gradient distribution similarity, hot and cold interface motion pattern similarity, and cold energy generation rate similarity, to solve the problem of mismatch between the focus of traditional methods and cold energy recovery needs; a stage-by-stage similarity weighting strategy is constructed to dynamically adjust the weights of each feature in the similarity calculation according to the different stages of the cold energy recovery process (initial rapid cooling stage, stable recovery stage, and final inefficient recovery stage) to improve the identification accuracy of key stages; cold energy recovery efficiency correlation analysis is introduced to establish a mapping relationship between working condition similarity and cold energy recovery efficiency, so that the similarity calculation directly points to the goal of optimizing cold energy recovery efficiency.

[0118] Based on the operating condition similarity vector S, an adaptive method for generating cold energy recovery control strategies is constructed, outputting an optimized cold energy recovery control strategy O. Specifically, the method addresses the limitations of traditional control strategy generation methods based on similar operating conditions by constructing a cold energy recovery key factor extraction technique. This technique extracts control decision factors highly correlated with cold energy recovery efficiency from historical control experience under similar operating conditions, avoiding the suboptimal control caused by indiscriminate transfer in traditional methods. A throttling-recovery dual-objective optimization framework is constructed, treating throttling valve control and cold energy recovery device control as a coupled optimization problem to maximize cold energy recovery while meeting turbine inlet parameter requirements. Furthermore, a cold energy quality assessment mechanism is established to conduct a comprehensive evaluation based on the temperature level (quality) and quantity of cold energy, guiding the control strategy generation process to prioritize the recovery of high-quality cold energy.

[0119] The cooling energy recovery optimization control strategy O and the hierarchical control strategy P are combined to produce a final integrated control instruction set I, which includes control signals for the throttle valve and the cooling energy recovery device. Integrating these two strategies within a model predictive control framework ensures stable system operation while maximizing cooling energy recovery efficiency.

[0120] like Figure 3 As shown, according to one aspect of the present application, the step of obtaining the optimized operating condition similarity vector includes:

[0121] Based on the system basic data set and dynamic cooling energy distribution prediction map, the cooling energy characteristic indicators are extracted and the cooling energy feature vector set is constructed;

[0122] Based on the cold energy feature vector set, a similarity calculation formula for cold energy recovery is constructed through thermodynamic similarity indicators to generate a thermodynamic similarity indicator set;

[0123] According to the characteristics of the cooling energy recovery process, a stage-by-stage similarity weighting strategy is constructed to generate a stage-by-stage weighting function set.

[0124] Read the cold energy recovery device operating data in the system basic data set, analyze the relationship between operating condition similarity and cold energy recovery efficiency, and establish an efficiency correlation map;

[0125] Combining the thermodynamic similarity index set, stage weighting function set and efficiency association mapping, the enhanced similarity between the current and historical operating condition similarity vectors is calculated, and the optimized operating condition similarity vector is output.

[0126] According to one aspect of the present application, the set of thermodynamic similarity indicators includes:

[0127] The temperature gradient distribution similarity index is obtained by calculating the cosine similarity of the temperature gradient fields of two working conditions;

[0128] The similarity index of the motion pattern of the hot and cold interfaces is obtained by comparing the movement speed and direction of the hot and cold interfaces;

[0129] The similarity index of the cold energy generation rate is obtained by comparing the change pattern of the cold energy generation per unit time.

[0130] The cold energy recovery process is divided into the initial rapid cooling stage, the stable recovery stage and the final inefficient recovery stage. A specific weight function is designed for each stage so that the similarity calculation can automatically adjust the importance of each feature according to the current stage, forming a stage weighted function set.

[0131] Specifically, based on the system's comprehensive monitoring dataset D and the dynamic cold energy distribution prediction map G, cold energy characteristic indicators were extracted and a cold energy feature vector set F_cold was constructed. To address the problem that traditional operating condition characteristics cannot fully express cold energy characteristics, a cold energy feature extraction method was developed. First, the spatial temperature gradient distribution index was calculated to quantify the clarity and location of the hot and cold interfaces. Then, the time derivative distribution of the temperature field was analyzed to characterize the rate and stability of cold energy generation. Finally, the spatial distribution pattern of cold energy density and the trend of total amount were combined to form a multidimensional feature vector that comprehensively characterizes the cold energy characteristics.

[0132] A thermodynamic similarity index system was constructed, a similarity calculation formula for cold energy recovery was designed, and a thermodynamic similarity index set H_index was generated. Traditional operating condition similarity primarily focuses on general operating characteristics (such as pressure and flow). This embodiment introduces a characteristic dimension centered on cold energy: first, "temperature gradient distribution similarity" is defined to calculate the cosine similarity of the temperature gradient fields of two operating conditions; then, "hot and cold interface motion pattern similarity" is designed to compare the movement speed and direction of the hot and cold interfaces; finally, "cold energy generation rate similarity" is created to compare the changing pattern of cold energy generation per unit time. These indicators directly target the core characteristics of cold energy recovery, improving the targeted identification of similar operating conditions.

[0133] Based on the characteristics of the cooling energy recovery process, a phase-by-phase similarity weighting strategy was developed, generating a set of phase-weighted functions, W_phase. A dynamically adjusted weighting scheme was constructed to account for the differences in characteristics across the cooling energy recovery process. First, the cooling energy recovery process was divided into an initial rapid cooling phase, a stable recovery phase, and a final, inefficient recovery phase. Key influencing factors for each phase were then analyzed, such as the initial focus on temperature drop rate, the stable phase on heat exchange efficiency, and the final phase on energy recovery equilibrium. Finally, a specific weighting function was designed for each phase, enabling the similarity calculation to automatically adjust the importance of each feature based on the current phase.

[0134] Read the historical operating data of the cold energy recovery device, analyze the relationship between the similarity of the working conditions and the cold energy recovery efficiency, and establish the efficiency association map R_map. Directly associate the identification of similar working conditions with the recovery efficiency target: first, extract the correspondence between the working condition characteristics and the actual recovery efficiency from the historical data; then use the supervised learning method to train the mapping model of the working condition characteristics to the recovery efficiency; finally, based on the efficiency sensitive factors predicted by the model, further adjust the feature weights in the similarity calculation so that the similarity calculation results point more directly to the direction of the optimal cold energy recovery efficiency. Combined with the thermodynamic similarity index set H_index, the stage weighted function set W_phase and the efficiency association map R_map, the enhanced similarity between the current working condition and the historical working condition is calculated, and the working condition similarity vector S is output. This embodiment improves the accuracy of cold energy recovery working condition identification from 70% of the traditional method to more than 90%, providing a high-quality similar working condition reference for subsequent control strategy generation.

[0135] like Figure 4 As shown, according to one aspect of the present application, the steps of adaptively generating a cold energy recovery control strategy and outputting a cold energy recovery optimization control strategy include:

[0136] Based on the optimized operating condition similarity vector, control strategies for similar operating conditions are retrieved from the historical control database, key control decisions are extracted, and a candidate control strategy set is generated;

[0137] Identify the control decision factors related to cooling energy recovery efficiency from the candidate control strategy set and construct a cooling energy efficiency influencing factor set;

[0138] Based on the coupling influence model and the cooling energy efficiency influencing factor set, a throttling-recovery dual-objective optimization framework is constructed. The throttling valve control and the cooling energy recovery device control are regarded as coupled optimization problems, and a multi-objective optimization model is constructed.

[0139] The multi-objective optimization model is combined with the current system state to perform real-time optimization calculations to generate a cold energy recovery optimization control strategy under the current working conditions, which is used to merge with the hierarchical control strategy to form the comprehensive control instruction set.

[0140] Specifically, based on the operating condition similarity vector S, control strategies for similar operating conditions are retrieved from the historical control database, key control decisions are extracted, and a set of candidate control strategies, C_strategies, is generated. Unlike traditional methods that simply apply the control strategy for the most similar operating condition, this embodiment adopts a more refined strategy extraction method: first, the top N historical operating conditions ranked by similarity are selected; then, the control strategy for each operating condition is decomposed to identify key control decisions such as the throttle valve opening trajectory and the working fluid flow adjustment mode; finally, the effect of each decision in the historical scenario is evaluated, and control elements with a significant positive impact on the cooling energy recovery efficiency are retained to form a set of candidate strategies.

[0141] A technique for extracting key factors in cooling energy recovery was developed to identify control decision factors highly correlated with cooling energy recovery efficiency from a set of candidate control strategies, C_strategies, and construct a set of cooling energy efficiency influencing factors, E_factors. Control strategies were directly linked to cooling energy recovery efficiency. First, a fine-grained analysis of historical strategy implementation results was conducted to quantify the impact of different control parameters on cooling energy recovery efficiency. A sensitivity analysis was then used to identify key control parameters and their optimal ranges. Finally, a response surface model was constructed to correlate control parameters with recovery efficiency, providing a theoretical basis for subsequent optimization.

[0142] A dual-objective throttling and recovery optimization framework is constructed, treating throttling valve control and cold energy recovery device control as a coupled optimization problem, and constructing a multi-objective optimization model M_opt. Addressing the limitations of single-objective optimization in traditional methods, a collaborative optimization approach is developed: First, based on the coupling influence model C and the set of cooling energy efficiency influencing factors E_factors, a mathematical model is established to describe the relationship between throttling valve parameters, cold energy recovery device parameters, and system performance (power generation efficiency, cold energy recovery efficiency). A multi-objective function is then defined, encompassing turbine inlet parameter stability and cold energy recovery maximization. Finally, a Pareto-optimal solution algorithm is designed to maximize cold energy recovery while satisfying turbine inlet parameter requirements.

[0143] Based on the current system state and the multi-objective optimization model M_opt, real-time optimization calculations are performed to generate the optimal cold energy recovery control strategy O for the current operating conditions. The theoretical model is combined with the current actual operating conditions: first, the real-time system state is read to update the initial conditions and constraints in the optimization model; then, a model predictive control (MPC) calculation is performed to predict the effects of different control strategies over a period of time in the future; finally, a control sequence that maximizes cold energy recovery while meeting turbine operating requirements is selected to form a complete control strategy. This embodiment achieves a synergistic effect between throttling control and cold energy recovery through coupled optimization. Compared with traditional separate control, the cold energy recovery efficiency is increased by 20-30%, while ensuring the stable operation of the turbine system.

[0144] In a specific embodiment of the present application, the dynamic utilization of cold energy based on a million cubic level compressed air energy storage system is implemented. This embodiment is based on an underground rock salt cave gas storage reservoir with a capacity of 1.2×10 6 m 3 , the working pressure range is 4.5-8.0MPa, the system rated power is 100MW, and the designed deflation time is 10 hours. The specific steps are:

[0145] Step 1: Obtain multi-source monitoring data of the compressed air energy storage system and build a basic data set for the system.

[0146] S11. Collect temperature and pressure data at multiple points inside the gas storage reservoir and construct a spatially distributed monitoring data set.

[0147] 120 measuring points are arranged inside the gas storage, including 86 temperature sensors (T1, T2, ..., T 86 ) and 34 pressure sensors (P1, P2, ..., P 34 Each sensor position is represented by spatial coordinates (X, Y, Z), and the sampling frequency is 1 time / 10 seconds. Taking the temperature data at time t=0 during a gas release process as an example: T1(50, 30, 20) = 25.3℃ (coordinate unit: m, origin is at the bottom center of the gas storage); T2(50, 30, 40) = 24.8℃; T3(50, 30, 60) = 24.1℃...; T 86 (750, 500, 300) = 23.6°C; Pressure data example: P1(50, 50, 50) = 7.85MPa; P2(150, 150, 50) = 7.84MPa ...; P 34 (750, 750, 300) = 7.82MPa.

[0148] S12. Acquire throttle valve operating status data to form a throttle control data set.

[0149] The system is configured with three parallel throttle valves. The system collects data on their opening angles (Vo1, Vo2, Vo3), pressure differentials across the throttle valves (ΔP1, ΔP2, ΔP3), temperature differentials (ΔT1, ΔT2, ΔT3), and flow rates (Q1, Q2, Q3) at a sampling rate of 1 per second. For example, the initial data is as follows: Vo1 = 35%, ΔP1 = 3.42 MPa, ΔT1 = 68.5°C, Q1 = 48.3 kg / s; Vo2 = 30%, ΔP2 = 3.40 MPa, ΔT2 = 67.8°C, Q2 = 41.5 kg / s; Vo3 = 0%, ΔP3 = 0 MPa, ΔT3 = 0°C, Q3 = 0 kg / s.

[0150] S13. Monitor the operating parameters of the cold energy recovery device and establish a cold energy recovery characteristic data set.

[0151] The cold energy recovery device uses ethylene glycol-water solution as the working fluid. Parameters such as inlet temperature (Tci), outlet temperature (Tco), working fluid flow rate (Qc), and heat exchange efficiency (η) are collected at a sampling frequency of once every 5 seconds. Example of initial data: Tci = 0°C, Tco = -20.5°C, Qc = 25.0 kg / s, and η = 0.85.

[0152] S14. Integrate multi-source data and perform time alignment and cleaning to output a standardized system comprehensive monitoring data set D.

[0153] The data with different sampling frequencies are time-aligned, with 1 time / 10 seconds as the standard sampling frequency. The throttling control data set is subjected to average downsampling, and the missing data are supplemented by linear interpolation, finally forming the system comprehensive monitoring data set D.

[0154] Step 2: Establish a space-time dynamic model inside the gas storage reservoir to accurately characterize the cold energy distribution characteristics.

[0155] S21. Use the kernel function interpolation method to preliminarily reconstruct the temperature field and pressure field of the entire space.

[0156] The radial basis function (RBF) interpolation method is used to preliminarily reconstruct the full-space temperature field Ts(x, y, z, t) and pressure field Ps(x, y, z, t). The calculation method is as follows: The calculation formula for the preliminary reconstruction of the temperature field is: Ts(x, y, z, t) = ∑[i=1 to n] wi ×φ(||r - ri||); where φ(r) = (r 2 + c 2 ) 1 / 2 is the radial basis function; r is the target point coordinate vector (x, y, z); ri is the coordinate vector of the i-th sensor; wi is the weight coefficient; c is the shape parameter, which is 1.5 times the average sensor spacing, 25 meters in this example; and n is the number of sensors. The weight coefficient wi is obtained by solving the linear equation system so that the value of the interpolation function at the sensor location equals the measured value. At a specific time (t = 0), the values of the preliminary reconstructed temperature field at some grid points are: Ts(100, 100, 50, 0) = 24.6°C; Ts(200, 200, 100, 0) = 24.2°C; and Ts(300, 300, 150, 0) = 23.9°C.

[0157] S22. Apply the multi-scale thermal flow constraint space reconstruction method to optimize the temperature and pressure fields in the entire space.

[0158] S221. Apply the hierarchical spatial segmentation algorithm to divide the gas storage space into multi-scale domains. According to the temperature gradient, the gas storage space is divided into three scales: microscale (0.1-1m 3 ): Mainly distributed near the wall of the gas storage reservoir and the throttle valve outlet area; mesoscale (1-100m 3 ): Mainly distributed in areas with large vertical temperature gradients; macroscale (>100m 3): Areas with gentle temperature changes. The octree algorithm is used for spatial partitioning, with an initial grid size of 50m. The grid size is adaptively refined based on the temperature gradient. Areas with a temperature gradient greater than 0.1°C / m are refined to a 5m grid size.

[0159] S222. Construct physical constraint models for different scales to form a multi-scale physical constraint set. Apply different physical models to different scale regions: Apply the heat diffusion equation at the microscale: ΨT / Ψt = α ▽ 2 T; where T is temperature (°C); t is time (s); Ψ is the partial derivative; α is the thermal diffusion coefficient, which is approximately 2.2×10 -5 m 2 / s; ▽ 2 is the Laplace operator, ▽ 2 T = Ψ 2 T / Ψx 2 + Ψ 2 T / Ψy 2 + Ψ 2 T / Ψz 2 The convection heat transfer model is used at the mesoscale: ΨT / Ψt + u· ▽ T = α ▽ 2 T; where u is the velocity vector (m / s), estimated from the pressure gradient; ▽ T is the temperature gradient, ▽ T = (ΨT / Ψx, ΨT / Ψy, ΨT / Ψz). A laminar flow model is used at the macroscale. The vertical temperature distribution equation in this laminar flow model is: T(z) = Tb + (Tt - Tb) × [1 - exp(-βz)]; where Tb is the bottom temperature (°C); Tt is the top temperature (°C); z is the normalized height; and β is the stratification parameter, obtained by fitting measured data; in this example, β = 2.3. Continuity boundary conditions are set at scale boundaries to ensure continuity of the temperature field and heat flux density.

[0160] S223. Perform physics-guided interpolation reconstruction to optimize the temperature and pressure fields in the entire space. Discretize the physical constraint equations into a system of linear equations and jointly solve the interpolation problem: Physics-guided interpolation reconstruction: min‖Ax-b‖ 2 + λ‖Lx‖ 2Where A is the observation matrix, with dimensions n×m, where n is the number of sensors and m is the number of reconstruction grid points; x is the temperature field vector to be calculated; b is the sensor measured temperature vector; L is the physical constraint operator matrix, constructed based on the physical model of each region; and λ is the balance parameter, set to 0.15. The conjugate gradient method is used to solve the above least squares problem, resulting in the optimized temperature field. Compared to the initial reconstruction, the temperature reconstruction accuracy in the throttle valve outlet region is improved by approximately 12%.

[0161] S224. Build an adaptive boundary constraint mechanism and a dynamic boundary condition set. Use the gas storage wall temperature sensor data to build dynamic boundary conditions: -k ▽ T·n = h(Tw - T); where k is the thermal conductivity of the medium (W / m·K); ▽ T is the temperature gradient; n is the wall normal vector; h is the convective heat transfer coefficient (W / m 2 K), obtained by fitting historical data, is 5.8; Tw is the wall temperature (°C); and T is the fluid temperature (°C). Applying this boundary condition improves the accuracy of temperature field reconstruction near the wall from 85% to 96%.

[0162] S23. Construct a time-space coupled evolution prediction algorithm to generate a dynamic cold energy distribution prediction map.

[0163] S231. Apply adaptive time window analysis to identify the temporal dynamic characteristics of different stages. Calculate the statistical characteristics of the temperature change rate of the entire space: mean (μ): the average temperature change rate of each grid point; variance (σ 2 ) is the spatial discreteness of the temperature change rate; Kurtosis (κ) reflects the distribution of extreme temperature changes. The outgassing process is divided into three stages: rapid change stage (0-1.5h): μ = -0.52℃ / min, σ 2 = 0.125, κ = 3.8; Transition period (1.5-4h): μ = -0.28℃ / min, σ 2 = 0.063, κ = 2.6; Slow change stage (4-10h): μ = -0.11℃ / min, σ 2 = 0.024, κ = 2.1. Different prediction time steps are assigned according to the characteristics of each stage: fast change stage: time step is 30 seconds; transition stage: time step is 2 minutes; slow change stage: time step is 5 minutes.

[0164] S232. Develop a temperature gradient sensitive spatial evolution model to form a segmented spatial evolution model set. Set different model parameters and grid density for different temperature gradient regions: Temperature gradient sensitive weight function: w( ▽ T) = 1 + α ×|▽ T| 2 ;Where w is the weight factor; ▽ T is the temperature gradient; α is the sensitivity parameter, which is 100 (℃ / m) -2 . According to the weight factor, the spatial evolution model is adjusted: in high gradient areas (such as near the vent, | ▽ T| > 0.2℃ / m): use a fine grid (mesh size 3m) and a fully coupled convection model; in the moderate gradient region (0.05℃ / m < | ▽ T| < 0.2℃ / m): Use medium grid (grid size 10m) and simplified convection model; in low gradient areas (| ▽ T| < 0.05°C / m): Use a coarse grid (mesh size 30m) and a thermal diffusion model.

[0165] S233. Analyze the impact of changes in throttle valve opening on the evolution of the temperature field inside the gas storage reservoir and establish a degassing-temperature coupling model. Throttle valve dynamic model: ΔT = f(P*1, P*2, Vo); where ΔT is the throttle temperature drop (°C); P*1 is the pressure before throttling (MPa); P*2 is the pressure after throttling (MPa); Vo is the valve opening (%); f is the coupling function, obtained by fitting experimental data. In this embodiment, the throttle temperature drop calculation formula is: ΔT = k × (P*1 – P*2) × [1 - 0.08 × (1 -Vo / 100) 2 ]; k is the proportionality coefficient, approximately 20°C / MPa; and the throttle valve opening Vo ranges from 0 to 100%. Based on this model, when the throttle valve opening is adjusted from 35% to 40%, the predicted temperature drop varies from 65.2°C to 62.8°C, which is less than 2°C different from the measured value.

[0166] S234. Use a multi-step progressive prediction algorithm to generate a dynamic cold energy distribution prediction map. Couple the temperature field prediction, exhaust flow and temperature drop model to perform a multi-step progressive prediction: Based on the current temperature field and throttling parameters, predict the temperature distribution of the next time step; update the boundary conditions and internal heat sources; use the prediction results as the initial state of the next step and continue the prediction. Cold energy density calculation: E_cold(x, y, z, t) = ρ × cp × max(0, T_amb - T(x, y, z, t)); where E_cold is the cold energy density (J / m 3 ); ρ is the air density (kg / m 3); cp is the specific heat capacity of air at constant pressure (J / kg·K); T_amb is the ambient reference temperature, which is 20°C; and T(x, y, z, t) is the predicted temperature field. The generated dynamic cooling energy distribution prediction map shows that cooling energy is mainly concentrated near the throttle valve outlet and the top of the reservoir, with the maximum cooling energy density reaching 8.5×10 5 J / m 3 The average error between the predicted cold energy distribution and the actual measured value is 7.6%, which is better than the 18.5% error of the traditional method.

[0167] Step 3: Develop a multi-time-scale hierarchical control framework to achieve coordinated control of the throttle valve and the cold energy recovery device.

[0168] S31. Establish a coupling impact model of throttling and cold energy recovery.

[0169] S311. Generate a coupled analysis data set. Time-align and preprocess the throttle valve operating state series and the cold energy recovery device operating data: normalize the data to the interval [-1, 1]. Use a sliding window method (10-minute window width, 1-minute step length) to segment the time series. Use the 3σ criterion to remove outliers.

[0170] S312. Use Granger causality analysis to identify causal relationships. Establish a causal relationship between throttle valve parameters and cold energy recovery performance: Granger Causality Index (GCI) (X→Y) = ln[var(ε_r) / var(ε_ur)]; where GCI is the Granger Causality Index (GCI); var(ε_r) is the prediction error variance of the restricted model; var(ε_ur) is the prediction error variance of the full model; X is the dependent variable, such as throttle valve opening; and Y is the outcome variable, such as cold energy recovery efficiency. The analysis results show that the Granger Causality Index (GCI) for throttle valve opening and cold energy recovery inlet temperature is 2.36, indicating a significant causal relationship between changes in throttle valve opening and changes in inlet temperature after 3-5 minutes.

[0171] S313. Construct a coupled impact model for throttling and cold energy recovery. Based on the results of causal analysis, a state-space model was established: State-space model: X(t+1) = A·X(t) + B·U(t), Y(t) = C·X(t) + D·U(t); where X is the state vector, including cold energy density and temperature; U is the control vector, including throttle valve opening and working fluid flow; Y is the output vector, including recovery efficiency and recovery power; A, B, C, and D are system matrices obtained through historical data identification. Model validation shows that the coupled model has a 92% prediction accuracy for cold energy recovery efficiency, a 15% improvement over traditional independent models.

[0172] S32. Build a hierarchical control framework and generate a hierarchical control strategy.

[0173] Based on the varying timescales of the system's dynamic characteristics, a three-layer control framework was designed: the strategy layer (hourly timescale) is responsible for overall operational strategy planning; the coordination layer (minutely timescale) coordinates throttling and cooling energy recovery; and the execution layer (seconds timescale) implements basic PID control functions. The output of each controller layer serves as the input or constraint for the controllers below it, forming a hierarchical control strategy.

[0174] S33. Build a collaborative control algorithm and output a control signal.

[0175] A model predictive control (MPC) approach is used to implement coordinated control: the objective function is: J = w1·J_throttle + w2·J_recovery; where J_throttle is the throttling control objective function, ensuring stable turbine inlet parameters; J_recovery is the cooling energy recovery objective function, maximizing cooling energy recovery. w1 and w2 are weight coefficients, dynamically adjusted according to the operating stage: w1 = 0.7, w2 = 0.3 in the initial stage, w1 = w2 = 0.5 in the intermediate stage, and w1 = 0.3, w2 = 0.7 in the later stage. Constraints include: turbine inlet temperature range: -5°C ≤ T_in ≤ 5°C; turbine inlet pressure fluctuation: |dP / dt| ≤ 0.05 MPa / min; working fluid flow rate range: 10 kg / s ≤ Qc ≤ 40 kg / s; and cooling energy recovery device safety constraint: Tco ≥ -35°C. By solving this optimization problem, the optimal throttling valve control signal and working fluid flow control signal are obtained.

[0176] Step 4: Construct a similar working condition identification and control strategy generation method to optimize the cooling energy recovery efficiency under different working conditions.

[0177] S41. Construct a fingerprint library of compressed air energy storage system operating conditions.

[0178] Based on historical operating data, principal component analysis (PCA) and clustering methods were applied to identify eight typical operating conditions: high-pressure rapid deflation; high-pressure stable deflation; medium-pressure rapid deflation; medium-pressure stable deflation; low-pressure rapid deflation; low-pressure stable deflation; pressure fluctuation deflation; and terminal low-pressure deflation. For each operating condition, a feature vector containing 20 characteristic parameters was extracted to form a fingerprint library of operating condition characteristics.

[0179] S42. Develop a working condition identification algorithm to generate a working condition similarity vector.

[0180] The cosine similarity method is used to calculate the similarity between the current operating condition and typical operating conditions in the database: Cosine similarity calculation: sim(A, B) = (A·B) / (||A||·||B||); where A is the characteristic vector of the current operating condition; B is the characteristic vector of the typical operating condition; · represents the vector dot product; ||A|| represents the Euclidean norm of vector A. For a real-time operating condition, the calculated similarity vector S = [0.78, 0.92, 0.45, 0.33, 0.26, 0.21, 0.15, 0.09] indicates that the current operating condition is most similar to the second type of "high-pressure stable bleed condition."

[0181] S43. Design a method to enhance the similarity of working conditions based on cooling energy characteristics.

[0182] S431. Extract cold energy characteristic indicators and construct a cold energy feature vector set. From the system data and cold energy distribution prediction map, extract the following cold energy characteristic indicators: spatial temperature gradient distribution indicator: μ_grad = 0.15°C / m, σ_grad = 0.04°C / m; hot and cold interface clarity: CI = 0.83 (range 0-1, larger values indicate clearer interfaces); cold energy generation rate: CR = 2.8 MW; cold energy density spatial distribution pattern: mainly concentrated in the top and outlet areas; total cold energy change trend: steadily increasing at a rate of approximately 1.5 MJ / s. These indicators constitute the cold energy feature vector F_cold.

[0183] S432. Develop a thermodynamic similarity index system. Construct a similarity calculation formula for cold energy recovery: Temperature gradient distribution similarity: Sim_grad(A, B) = ( ▽ TA· ▽ TB) / (|| ▽ TA||·|| ▽ TB||); where ▽ TA, ▽ TB is the temperature gradient field of working conditions A and B. Similarity of hot and cold interface motion mode: Sim_interface(A, B) = exp(-||VA - VB|| 2 / σ 2 ); where VA and VB are the hot and cold interface velocity vectors for conditions A and B; σ is a normalization parameter, set to 0.5 m / s. Cold energy generation rate similarity: Sim_rate(A, B) = 1 - |CRA - CRB| / max(CRA, CRB); where CRA and CRB are the cold energy generation rates for conditions A and B.

[0184] S433. Develop a phased similarity weighting strategy. Divide the cold energy recovery process into three phases and construct different weighting functions: Initial rapid cooling phase (t < 1.5h): W_init(t) = [0.5, 0.15, 0.35]; weights correspond to temperature gradient, interface motion, and generation rate similarity, respectively. Stable recovery phase (1.5h ≤ t < 8h): W_stable(t) = [0.2, 0.3, 0.5]; emphasis is placed on cold energy generation rate similarity. Final inefficient recovery phase (t ≥ 8h): W_end(t) = [0.15, 0.5, 0.35]; emphasis is placed on interface motion pattern similarity.

[0185] S434. Establish an efficiency association mapping and calculate enhanced similarity. Analyze the relationship between historical operating conditions and recovery efficiency and establish a mapping model: Efficiency prediction model: η_pred = Σ[i=1 to 3] βi · Simi(A, B); where η_pred is the predicted recovery efficiency; Simi is the i-th similarity index; and βi is the regression coefficient, obtained through training with historical data and set to [0.35, 0.25, 0.4]. Based on this model, adjust the final weights for the similarity calculation and output the operating condition similarity vector S. Practice has shown that the enhanced operating condition recognition accuracy has increased from 70% to 91%.

[0186] S44. Develop an adaptive generation method for cold energy recovery control strategy.

[0187] S441. Retrieve control strategies for similar operating conditions and generate a candidate control strategy set. Retrieve the top five similarity-ranked operating conditions from the historical control database and extract their control strategies: throttle valve opening trajectory: Vo(t) = f1(t); working fluid flow rate regulation mode: Qc(t) = f2(t); working fluid temperature set point: Tset(t) = f3(t).

[0188] S442. Extract control decision factors that are highly correlated with cold energy recovery efficiency. Identify key control parameters through sensitivity analysis: Sensitivity calculation: S(η, p) = (Ψη / Ψp)·(p / η); where S is the sensitivity; η is the cold energy recovery efficiency; p is the control parameter; and Ψη / Ψp is the partial derivative of the efficiency with respect to the parameter. The analysis results show that the working fluid flow rate has a sensitivity of 0.78 to the recovery efficiency, making it the most critical parameter; the throttle valve opening rate has a sensitivity of 0.65, making it the second most critical parameter.

[0189] S443. Construct a dual-objective optimization framework for throttling and recovery. Construct a multi-objective optimization model: Objective function: min[-w1·η_recovery - w2·s_stability]; where η_recovery is the cooling energy recovery efficiency, ranging from 0 to 1; s_stability is the turbine inlet parameter stability index, ranging from 0 to 1; w1 and w2 are weight coefficients, set to 0.6 and 0.4, respectively. Constraints include technical and safety constraints, such as valve actuation speed limits and working fluid temperature safety limits.

[0190] S444. Perform real-time optimization calculations to generate an optimized control strategy for cold energy recovery. A genetic algorithm is used to solve the multi-objective optimization problem, obtaining a Pareto-optimal solution set. The most appropriate control strategy is then selected based on the current operating conditions. The resulting optimized control strategy for cold energy recovery includes: a throttle valve opening adjustment curve; a cold energy recovery fluid flow rate adjustment curve; and a fluid temperature setpoint curve.

[0191] S45, integrate control strategies and output comprehensive control instruction sets.

[0192] The hierarchical control strategy and the cooling energy recovery optimization control strategy are integrated to generate a comprehensive control instruction set. When the objectives are consistent, the respective strategies are directly adopted. When the objectives conflict, a weighted average method is used to resolve the conflict: U_final = α·U_layer + (1-α)·U_recovery, where U_final is the final control instruction; U_layer is the output of the hierarchical control strategy; and U_recovery is the output of the cooling energy recovery optimization control strategy. α is the weight coefficient, which defaults to 0.5. The comprehensive control instruction set outputs the throttle valve control signal and the cooling energy recovery device control signal at a frequency of 1 second.

[0193] This embodiment has been validated in an actual million-cubic-meter compressed air energy storage system, achieving significant results compared to traditional methods. The average error in temperature field reconstruction accuracy using a traditional pure interpolation method is 18.5%, while this embodiment reduces the average error to 7.6%. Reconstruction accuracy is particularly improved by over 40% in areas with sparse measurement points. The cooling energy recovery efficiency is 62% under the traditional independent control method, but this embodiment increases it to 81% through the use of coordinated control, a 19 percentage point improvement. The turbine inlet temperature fluctuation is ±4.2°C under the traditional method, but this embodiment reduces it to ±1.8°C, improving stability by 57%. The system cycle efficiency is 55.6% under the traditional method, but this embodiment increases it to 59.8% through efficient cooling energy recovery, a relative improvement of 7.5%. The accuracy of operating condition identification based on traditional general features is 70%, while this embodiment achieves 91% accuracy based on cooling energy characteristics. Over a 10-hour deflation cycle, the average deviation of cooling energy location prediction using the traditional method is 35 meters, while this embodiment reduces it to 12 meters, improving positioning accuracy by 65%.

[0194] The following shows a comparison of the cold energy recovery efficiency during a deflation process: when the deflation period is 0-2 hours, the efficiency of the traditional method is 58.5%; the efficiency of this embodiment is 76.2%, with an improvement of 17.7%; when the deflation period is 2-5 hours, the efficiency of the traditional method is 64.3%; the efficiency of this embodiment is 85.7%, with an improvement of 21.4%; when the deflation period is 5-8 hours, the efficiency of the traditional method is 66.1%; the efficiency of this embodiment is 83.8%, with an improvement of 17.7%; when the deflation period is 8-10 hours, the efficiency of the traditional method is 55.2%; the efficiency of this embodiment is 72.6%, with an improvement of 17.4%; the average efficiency of the traditional method is 62.0%; the average efficiency of this embodiment is 81.0%, with an evaluation improvement of 19.0%;

[0195] According to one aspect of the present application, the technical implementation of the multi-scale thermal flow constraint space reconstruction method is as follows: In actual applications, there are only 120 measuring points arranged in a million-cubic-meter gas storage reservoir, with an average of only one measuring point per 10,000 cubic meters, and the spatial measurement points are extremely sparsely distributed. Traditional interpolation methods are difficult to accurately reconstruct the temperature distribution of such a large space, especially in areas with large temperature gradients. Multi-scale physical constraints are introduced, and the detailed implementation steps are as follows: Octree space segmentation method: The space is divided by recursive octant method, the initial cube side length is 100m, and it is adaptively subdivided according to the temperature gradient. The subdivision condition is: if the temperature gradient | ▽ T| > 0.2℃ / m, then subdivide to a side length of 5m; if 0.05℃ / m < | ▽ T| < 0.2℃ / m, then subdivide to a side length of 20m; if | ▽If T| < 0.05℃ / m, a large-scale grid (side length ≥ 50m) is maintained. Multi-physics model application method: physical models of different complexity are used for different scale regions, and the calculation accuracy and efficiency are dynamically balanced during the calculation process. The heat flow constraint is solved using the finite difference method, and the discretization format is: The discretization format of the microscale heat diffusion equation is: (T n+1_i，j，k - T n_i，j，k ) / Δt = α[(T n_i+1，j，k + T n_i-1，j，k - 2T n_i，j，k ) / (Δx) 2 + (T n_i，j+1，k + T n_i，j-1，k - 2T n_i，j，k ) / (Δy) 2 + (T n_i，j，k+1 + T n_i，j，k-1 - 2T n _i, j, k) / (Δz) 2 ]; where T n _i，j，k is the temperature at the spatial point (i, j, k) at time step n; Δt is the time step; Δx, Δy, and Δz are the spatial steps; and α is the thermal diffusion coefficient. The core algorithm for physics-guided interpolation: Construct a global objective function that takes physical constraints into account: J(T) = w1·J_data(T) + w2·J_phys(T) + w3·J_smooth(T); where J_data is the data fitting term, J_data(T) = Σ||T(x_i) - T_i|| 2 J_phys is the physical constraint term, constructed using different physical equations for different regions; J_smooth is the smoothing regularization term; w1, w2, and w3 are weight coefficients, which in this example are [0.6, 0.3, 0.1]. By solving this optimization problem, a temperature field that satisfies the physical constraints and is consistent with the measured data is obtained. This embodiment improves reconstruction accuracy by 40% in areas with temperature gradients greater than 0.1°C / m.

[0196] The specific implementation of the time-space coupled evolution prediction algorithm is as follows: Traditional prediction methods often use a uniform time step, which makes it difficult to balance computational efficiency and prediction accuracy. This embodiment constructs a non-uniform time step mechanism, which is specifically implemented as follows: Adaptive time step determination method: The time step is automatically adjusted according to the temperature change rate. The calculation formula is: Δt_adaptive = Δt_base × min(1, (|dT / dt|_threshold / |dT / dt|_max) β); where Δt_adaptive is the adaptive time step; Δt_base is the base time step, set to 300 seconds; |dT / dt|_threshold is the threshold temperature change rate, set to 0.2°C / min; |dT / dt|_max is the current maximum temperature change rate; β is the adjustment coefficient, set to 0.5. The specific implementation of the temperature gradient sensitive model: construct an adaptive weight coefficient and use a more refined calculation model in high temperature gradient areas. The weight coefficient calculation formula is: W_grad(i, j, k) = 1 + γ·(| ▽ T|_i,j,k / | ▽ T|_avg) Δ ;W_grad is the gradient weight coefficient;| ▽ T|_i, j, k is the temperature gradient at point (i, j, k); ▽ T|_avg is the average temperature gradient; γ is the amplification factor, set to 2.0; Δ is the exponential coefficient, set to 1.5. Accurate modeling of the degassing-temperature coupling model: Establish the coupling relationship between the throttling degassing process and the temperature field, using the principles of conservation of mass and energy. The specific formula is: dT / dt = -v· ▽ T + α· ▽ 2 T + S_T; where v is the velocity field, which is calculated from the bleed flow rate and pressure gradient; S_T is the temperature source term, which is related to the temperature drop caused by the throttle valve. The formula for calculating the temperature drop of the throttle valve is: ΔT_throttle = k·(P*1 ((κ-1) / κ) –P*2 ((κ-1) / κ) )·(P*1 / P*2); where ΔT_throttle is the throttling temperature drop; P*1 and P*2 are the pressures before and after throttling, respectively; k is the coefficient, approximately 358K; and κ is the air adiabatic index, approximately 1.4. The combined application of these technologies has significantly improved the accuracy of cooling energy distribution prediction maps, reducing the average prediction error from 18.5% with traditional methods to 7.6%.

[0197] The specific implementation of the working condition similarity enhancement method based on cooling energy characteristics is as follows: Traditional working condition identification methods are mainly based on general characteristics such as pressure and flow, and pay insufficient attention to cooling energy characteristics. This embodiment constructs a similarity calculation method specifically for cooling energy characteristics: The specific method of cooling energy feature extraction is to extract the following cooling energy features from the temperature field data: Temperature gradient vector field: ▽ T = (ΨT / Ψx, ΨT / Ψy, ΨT / Ψz); hot and cold interface position: defined by the temperature isosurface T = T_amb; hot and cold interface clarity: CI = | ▽ T|_interface / | ▽ T|_avg; where| ▽T|_interface is the temperature gradient at the hot and cold interface, | ▽ T|_avg is the average temperature gradient across the entire temperature field. The spatiotemporal distribution pattern of cold energy is determined by principal component analysis (PCA) to extract eigenvectors. Accurate calculation of thermodynamic similarity indices: Taking into account both the static and dynamic characteristics of the temperature field, similarity is calculated as follows: S_thermo = w1·S_gradient + w2·S_interface + w3·S_rate + w4·S_pattern; where S_gradient represents the temperature gradient similarity; S_interface represents the interface property similarity; S_rate represents the cold energy generation rate similarity; and S_pattern represents the distribution pattern similarity. w1, w2, w3, and w4 are weight coefficients, which are dynamically adjusted based on the system phase. The specific algorithm for the phased weighting strategy: Define a time-dependent weight function and automatically adjust the weights based on the system phase: W(t) = W_init·f_init(t) + W_stable·f_stable(t) + W_end·f_end(t); where W_init, W_stable, and W_end are the weight vectors for the initial, stable, and final phases, respectively; f_init, f_stable, and f_end are time-dependent phase functions, satisfying f_init(t) + f_stable(t) + f_end(t) = 1. This embodiment enables more accurate operating condition identification and more targeted recovery strategies for similar operating conditions, which is a key factor in improving cold energy recovery efficiency.

[0198] The specific implementation of the adaptive generation method of the cold energy recovery control strategy is as follows: This embodiment breaks through the limitations of traditional independent control strategies and establishes a coordinated control mechanism for throttling and cold energy recovery: Cold energy recovery key factor extraction technology: Using data mining methods, key factors affecting cold energy recovery efficiency are extracted from historical data: the matching degree of working fluid flow and throttling temperature drop: MD = Qc / (ΔT·Q_throttle), the optimal value is 0.12-0.15, where Q_throttle is the heat flow rate during the throttling process; the inlet temperature gradient of the cold energy recovery device: |dTci / dt|, the optimal range is < 0.8℃ / min; the working fluid flow adjustment rate: |dQc / dt|, the optimal range is < 0.5kg / s·min. Throttling-Recovery Dual-Objective Optimization Framework: An optimization model is constructed that comprehensively considers turbine stability and cooling energy recovery efficiency: f_obj = -[α·η_recovery + (1-α)·(1-σT / σT_max)]; where η_recovery is the cooling energy recovery efficiency; σT is the standard deviation of turbine inlet temperature fluctuation; σT_max is the maximum allowable standard deviation; and α is a trade-off coefficient, which defaults to 0.6. Cooling Energy Quality Assessment Mechanism: A temperature level is introduced to evaluate cooling energy quality, using the following formula: Q_cold = m_cold·cp·(T_ref - T_cold)·(1 + β·(T_ref - T_cold) / T_ref); where Q_cold is the cooling energy value considered for quality; m_cold is the mass flow rate of the cooling medium; cp is the specific heat capacity; T_ref is the reference temperature; T_cold is the cooling energy temperature; and β is the quality correction factor, which is set to 0.15. This embodiment enables the system to more effectively utilize the cold energy resources in the gas storage reservoir and realizes coordinated control with the turbine power generation process, effectively improving the overall system efficiency.

[0199] This embodiment details the specific implementation process of the method for dynamically utilizing cold energy in a million-cubic-meter compressed air energy storage system. It addresses technical challenges such as accurate modeling of the temperature field within large gas storage reservoirs, identification of cold energy characteristic operating conditions, and coordinated control of throttling and cold energy recovery, improving cold energy recovery efficiency while ensuring stable operation of the turbine system. Experimental verification shows that compared to traditional technologies, this embodiment improves cold energy recovery efficiency by 19 percentage points, improves turbine inlet parameter stability by 57%, and increases overall system efficiency by 7.5%, providing important technical support for the large-scale application of compressed air energy storage systems.

[0200] This invention divides the million-cubic-meter gas storage reservoir into three scale levels: micro, meso, and macro. An octree partitioning method is used to create an adaptive grid at each level. Different physical models are applied at different scales (a microscale heat diffusion equation, a mesoscale convective heat transfer model, and a macroscale laminar flow model). This approach addresses the "false distribution" problem inherent in traditional spatial reconstruction methods, which results from sparse sampling points in large gas storage reservoirs. Through physics-guided interpolation reconstruction and an adaptive boundary constraint mechanism, the accuracy of temperature field reconstruction is improved from 85% to over 95% in traditional methods, particularly in areas with sparse measurement points, with a 40% improvement. This provides precise spatial positioning support for cold energy recovery, ensuring that cold energy recovery devices can be effectively deployed and controlled based on actual cold energy distribution, avoiding energy waste and inefficient equipment operation. By developing a non-uniform time step prediction mechanism, a temperature gradient-sensitive spatial evolution model, and coupled analysis of venting flow and temperature field, accurate prediction of the dynamic evolution of the temperature field within the gas storage reservoir is achieved. Based on the statistical characteristics of the temperature change rate, this method automatically identifies rapid, transitional, and slow-changing phases, assigning appropriate time steps and particularly enhancing prediction accuracy for low-temperature areas and regions with large temperature gradients. By treating the throttling process and the evolution of the internal temperature field as a coupled system, it accurately captures the changes in cooling energy distribution under different venting strategies. This method reduces prediction error from 15-20% in traditional methods to 5-8%, particularly improving the accuracy of location prediction for cooling energy concentration areas by over 40%. This provides a reliable basis for feedforward control of the cooling energy recovery device, enabling advance planning and dynamic adjustment. By developing a thermodynamic similarity index system (temperature gradient distribution similarity, hot-cold interface motion pattern similarity, and cooling energy generation rate similarity) and a stage-by-stage similarity weighting strategy, the calculation of operating condition similarity is directly linked to the cooling energy recovery efficiency target. A similarity calculation formula is designed specifically for cooling energy characteristics, dynamically adjusting the weights of each feature according to the different stages of the cooling energy recovery process (initial rapid cooling, stable recovery, and final inefficient recovery). By introducing a cooling energy recovery efficiency correlation mapping, the similarity calculation is directly aligned with the cooling energy recovery efficiency optimization target. The accuracy of cold energy recovery condition identification has been increased from 70% in traditional methods to over 90%, providing a high-quality reference for similar operating conditions for subsequent control strategy generation, improving cold energy recovery efficiency and system operational stability, and avoiding the inefficient control problem caused by traditional methods that only focus on general operating characteristics. By developing key factor extraction technology for cold energy recovery, a throttling-recovery dual-objective optimization framework, and a cold energy quality assessment mechanism, the adaptive generation of cold energy recovery control strategies has been achieved. Control decision factors highly correlated with cold energy recovery efficiency are extracted from historical control experience of similar operating conditions. Throttle valve control and cold energy recovery device control are treated as a coupled optimization problem to maximize cold energy recovery while meeting turbine inlet parameter requirements.By establishing a cooling energy quality assessment mechanism, a comprehensive evaluation based on the temperature level (quality) and quantity of cooling energy is conducted, guiding the control strategy to prioritize the recovery of high-quality cooling energy. A coupled optimization approach achieves a synergistic effect between throttling control and cooling energy recovery, improving cooling energy recovery efficiency by 20-30% compared to traditional separate control. This approach also ensures stable operation of the turbine system, resolves the conflict between cooling energy recovery and turbine stability, and improves the overall energy efficiency of the system. Granger causal analysis is employed to identify the causal relationship between throttling valve parameters and cooling energy recovery performance, and a coupled influence model is established. First, a time-aligned coupled analysis dataset is extracted from multi-source data. Time lag analysis is then used to determine the causal direction between variables. A directed weighted network is constructed, and the identified key causal paths are quantified into a mathematical model. This model overcomes the limitations of traditional independent control and provides a theoretical basis for the coordinated control of throttling valves and cooling energy recovery devices. Model validation demonstrates that the coupled model achieves a 92% accuracy in predicting cold energy recovery efficiency, a 15% improvement over traditional standalone models. By understanding how changes in throttling parameters affect cold energy generation and distribution, the system can predictively adjust its control strategy, avoiding the cold energy waste and system fluctuations caused by control lag in traditional approaches. A three-layer control framework, consisting of a strategy layer (hourly), a coordination layer (minute-level), and an execution layer (second-level), achieves coordinated control of the throttling valve and the cold energy recovery device. This framework rationally allocates control tasks based on the varying timescales of the system's dynamic characteristics, with the outputs of the upper-layer controllers serving as inputs or constraints for the lower-layer controllers. Coordinated control is achieved using a model predictive control approach, dynamically adjusting weight coefficients to balance throttling control objectives with cold energy recovery objectives at different operating stages. This hierarchical control framework improves system response, reducing turbine inlet temperature fluctuations from ±4.2°C (compared to the traditional approach) to ±1.8°C, while increasing stability by 57%. This framework leverages the system's predictability and controllability, addresses time lag and stability issues in the cold energy recovery process, and enhances the system's resilience and adaptability to disturbances.

[0201] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the scope of protection of the present invention.

Claims

1. A method for dynamically utilizing cold energy from a million cubic meter compressed air energy storage system, characterized in that: include: Acquire multi-source monitoring data of the compressed air energy storage system to form a basic data set of the system; Based on this, a multi-scale heat flow constraint space reconstruction is carried out, and a dynamic cold energy distribution prediction map is generated through a space-time dynamic model; Based on the basic data set of the system, a hierarchical control framework is constructed through the coupling impact model of throttling and cold energy recovery to generate a hierarchical control strategy; Based on the system basic data set and dynamic cooling energy distribution prediction map, a working condition feature fingerprint library is constructed, and a cooling energy recovery optimization control strategy is generated by identifying similar working conditions; Integrate it with the hierarchical control strategy to output a comprehensive control instruction set; Generate an optimized control strategy for cold energy recovery, including: Based on the basic system data set, a working condition feature fingerprint library of the compressed air energy storage system is constructed; based on this, a working condition similarity vector is generated from the current system state through working condition identification; Combined with the dynamic cooling energy distribution prediction map, the operating condition similarity vector is enhanced based on the cooling energy characteristics to obtain the optimized operating condition similarity vector; based on this, the cooling energy recovery control strategy is adaptively generated and the cooling energy recovery optimization control strategy is output; The process of establishing the coupling impact model of throttling and cold energy recovery includes: Read the throttle valve operating state sequence and the cold energy recovery device operating data to generate a coupled analysis data set; Based on the coupled analysis data set, the Granger causal analysis method is used to identify the causal relationship between throttle valve parameters and cold energy recovery performance, and a dynamic causal network is constructed. Based on the dynamic causal network, a coupling impact model of energy saving and cooling energy recovery is constructed; Generate a hierarchical control strategy including: Based on the different time scales of the system's dynamic characteristics, a three-layer control framework is designed: the strategy layer is responsible for overall operation strategy planning; the coordination layer coordinates throttling and cold energy recovery; the execution layer implements basic PID control functions; the output of each layer controller serves as the input or constraint of the lower layer controller, forming a hierarchical control strategy.

2. The method according to claim 1, characterized in that Generate dynamic cooling energy distribution prediction map, including: Based on the system basic data set, the full-space temperature and pressure fields in the gas storage are preliminarily reconstructed. Based on this, multi-scale thermal flow constraint space reconstruction is performed to optimize the full-space temperature and pressure fields. Based on the optimized full-space temperature and pressure fields, time-space coupled evolution prediction is performed to generate a dynamic cold energy distribution prediction map.

3. The method according to claim 2, characterized in that Optimize the temperature and pressure fields of the entire space, including: The preliminary reconstruction results of the full-space temperature field and pressure field are subjected to hierarchical spatial segmentation to obtain a multi-scale spatial domain; based on this, physical constraint models for different scales are constructed to form a multi-scale physical constraint set; The gas storage wall temperature sensor data in the system basic data set is used to construct a dynamic boundary condition set through adaptive boundary constraints. This is combined with the sensor measured data and multi-scale physical constraint set in the system basic data set to perform physics-guided interpolation reconstruction and optimize the full-space temperature and pressure fields.

4. The method according to claim 2, characterized in that Conduct time-space coupled evolution prediction and generate dynamic cold energy distribution prediction maps, including: Based on the optimized full-space temperature and pressure fields, adaptive time window analysis is applied to identify the temporal dynamic characteristics of different stages and generate a set of temporal characteristic segments. For each stage, a temperature gradient-sensitive spatial evolution model is constructed to form a set of segmented spatial evolution models. The bleed parameters of the throttle valve operating state sequence in the system basic data set are read, the impact of throttle valve opening changes on the evolution of the temperature field inside the gas storage reservoir is analyzed, and a bleed-temperature coupling model is established. This model is combined with the segmented spatial evolution model set to generate a dynamic cold energy distribution prediction map through multi-step progressive prediction.

5. The method according to claim 4, characterized in that Generate a temporal feature segment set, including: Based on the optimized full-space temperature and pressure fields, the statistical characteristics of the full-space temperature change rate, including mean, variance, and kurtosis, are calculated. Through time series clustering, statistical features are classified to automatically identify rapid change, transition and slow change stages; the time step is adjusted according to the characteristics of different stages: the first small time step is used to improve prediction accuracy in the rapid change stage, and the second time step is used to improve calculation efficiency in the slow change stage; finally, a set of time characteristic segments is generated, in which the first time step is smaller than the second time step.

6. The method according to claim 1, wherein The optimized working condition similarity vector is obtained, including: Based on the system basic data set and dynamic cooling energy distribution prediction map, the cooling energy characteristic indicators are extracted, the similarity calculation formula for cooling energy recovery is constructed, and a thermodynamic similarity index set is generated; According to the characteristics of the cooling energy recovery process, a stage-by-stage similarity weighting strategy is constructed to generate a stage-by-stage weighting function set. Read the cold energy recovery device operating data in the system basic data set, analyze the relationship between operating condition similarity and cold energy recovery efficiency, and establish an efficiency correlation map; Combining the thermodynamic similarity index set, stage weighting function set and efficiency association mapping, the enhanced similarity between the current and historical operating condition similarity vectors is calculated, and the optimized operating condition similarity vector is output.

7. The method according to claim 6, characterized in that The set of thermodynamic similarity metrics includes: The temperature gradient distribution similarity index is obtained by calculating the cosine similarity of the temperature gradient fields of two working conditions; The similarity index of the motion pattern of the hot and cold interfaces is obtained by comparing the movement speed and direction of the hot and cold interfaces; The similarity index of the cold energy generation rate is obtained by comparing the change pattern of the cold energy generation per unit time.

8. The method according to claim 1, characterized in that Perform adaptive generation of cooling energy recovery control strategies and output cooling energy recovery optimization control strategies, including: Based on the optimized operating condition similarity vector, control strategies for similar operating conditions are retrieved from the historical control database, key control decisions are extracted, and a set of candidate control strategies is generated. From these strategies, control decision factors related to cooling energy recovery efficiency are identified, and a set of cooling energy efficiency influencing factors is constructed. Based on the coupling influence model and the set of cooling energy efficiency influencing factors, a throttling-recovery dual-objective optimization framework is constructed. The throttling valve control and the cooling energy recovery device control are regarded as coupled optimization problems, and a multi-objective optimization model is constructed. It is combined with the current system status to perform real-time optimization calculations and generate the cooling energy recovery optimization control strategy under the current operating conditions.

9. A million cubic meter compressed air energy storage system and a dynamic utilization system of cold energy in a gas storage tank, characterized by: include: Compression system, used to compress air; Gas storage, connected to the compression system, used to store compressed gas; An exhaust pipe connected to the gas storage reservoir and used to discharge the gas in the gas storage reservoir; A throttle valve is provided in the exhaust pipe to control the gas flow and pressure; A cold energy recovery device is provided at the rear end of the exhaust port of the throttle valve in the exhaust pipe and is used to recover cold energy in the gas; The heat exchanger is installed at the rear end of the cold energy recovery device and exchanges heat with the gas through cold energy; The turbine system is installed after the heat exchanger and is used to generate electricity by turning the gas after heat exchange; Control module, including: at least one processor; and, a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the method according to any one of claims 1 to 8.

Images