Compressed air energy storage system performance monitoring system construction method and monitoring system
By constructing a dual-driven hybrid model based on the combination of physical information and data, and combining deep neural networks with thermodynamic equations, we have achieved high-efficiency energy efficiency optimization and equipment life balance of compressed air energy storage systems in extreme environments, solved the shortcomings of traditional monitoring technologies in dynamic environmental changes and multivariable coupling modeling, and achieved high-precision and low-cost system operation.
Patent Information
- Application Number
- CN202510711611.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-29
- Publication Date
- 2025-09-16
AI Technical Summary
Existing monitoring technology for compressed air energy storage systems has deficiencies in environmental dynamics and multivariable coupling modeling, making it difficult to achieve a balance between efficient energy efficiency optimization and equipment life. In particular, the model's generalization ability is poor in extreme environments, leading to energy efficiency loss and equipment overload risks.
Build a dual-drive hybrid model based on the combination of physical information and data, combine deep neural networks with thermodynamic equations to perform self-optimization learning and real-time compensation, and combine edge computing with cloud-based collaborative learning to achieve millisecond-level response and high-precision prediction.
It significantly improves the accuracy of system status prediction and environmental adaptability, reduces energy consumption and operation and maintenance costs, extends equipment life, and ensures stable operation of the system in highly dynamic scenarios.
Smart Images
Figure CN120654545A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of energy storage and intelligent monitoring, and relates to a method for constructing a performance monitoring system and a monitoring system, and in particular to a method for constructing a performance monitoring system for a compressed air energy storage system and a monitoring system. Background Art
[0002] Performance monitoring of compressed air energy storage systems is a key component in ensuring energy storage efficiency and equipment lifespan. The key lies in real-time matching of dynamic environmental changes with system operating parameters. Traditional monitoring systems are mostly based on fixed thresholds or single variable adjustments, primarily relying on the following methods:
[0003] 1. Traditional threshold monitoring system: It triggers alarms or shutdown protection through preset pressure and temperature thresholds, but ignores the dynamic impact of environmental parameters (such as temperature, humidity, and air pressure) on the compression / energy release process, resulting in energy efficiency loss and equipment overload risks.
[0004] 2. Single variable compensation technology: Temperature, humidity, and pressure sensors are used to perform linear corrections on compressor power. Although this can alleviate the impact of local environment, it does not solve the nonlinear interference of multiple environmental variables on the thermodynamic characteristics of the gas, and the global energy efficiency optimization is insufficient.
[0005] 3. Static model control: PID or fuzzy control algorithms are trained based on historical data. However, the model has poor generalization capabilities in extreme environments (high temperature, high humidity) or real-time meteorological mutation scenarios.
[0006] In summary, existing technologies have significant defects in environmental dynamic adaptability, multivariable coupling modeling and real-time control accuracy, and it is difficult to balance the complex relationship between energy storage efficiency, equipment life and environmental disturbances. Summary of the Invention
[0007] In order to solve the above technical problems existing in the background technology, the present invention provides a method for constructing a compressed air energy storage system performance monitoring system and a monitoring system that can improve the accuracy of state prediction and enhance environmental adaptability.
[0008] In order to achieve the above object, the present invention adopts the following technical solutions:
[0009] A method for constructing a compressed air energy storage system performance monitoring system, characterized in that the method for constructing a compressed air energy storage system performance monitoring system comprises the following steps:
[0010] 1) Build a dual-driven hybrid model based on the combination of physical information and data;
[0011] 2) performing self-optimization learning on the dual-drive hybrid model combining the physical information and data obtained in step 1) to obtain a self-optimized dual-drive hybrid model combining the physical information and data;
[0012] 3) compensating the prediction result of the dual-drive hybrid model combining the self-optimized physical information and data obtained in step 2) based on the environmental variables to obtain a compensated dual-drive hybrid model combining the physical information and data;
[0013] 4) The dual-drive hybrid model combining the compensated physical information and data obtained in step 3) is trained using the data set to improve the output accuracy of the model and obtain a compressed air energy storage system performance monitoring system.
[0014] The specific implementation of step 1) above is:
[0015] 1.1) Selecting a gas state equation; preferably, the gas state equation is the ideal gas law or the van der Waals equation;
[0016] 1.2) Construct basic thermodynamic differential equations based on the gas state equation, the first law of thermodynamics, and the second law of thermodynamics;
[0017] 1.3) Embedding the basic thermodynamic differential equations obtained in step 1.2) into the training process of a deep neural network in the form of physical constraints to construct a dual-driven hybrid model that integrates physical information and data; preferably, the deep neural network is an LSTM or Transformer to adapt to modeling problems with time or state dependence characteristics; the physical information is information calculated by thermodynamic differential equations; the data is information supplementing the information calculated by thermodynamic equations, and the data includes simulation data, operating data of the compressed air energy storage system, data collected by sensors, and other experimental data.
[0018] The specific method of self-optimization learning in step 2) above is to update the weights of the deep neural network and / or improve the output accuracy;
[0019] The specific implementation method of updating the weights of the deep neural network is: real-time collection of operating data and environmental parameters of the compressed air energy storage system, and dynamic updating of the weights of the deep neural network in the dual-drive hybrid model combining physical information and data through an online gradient descent algorithm; preferably, the operating data includes pressure, temperature and power; the environmental parameters include temperature, humidity and air pressure;
[0020] The specific implementation method for improving the output accuracy is: introducing the physical equation residual term in the loss function to improve the output accuracy of the dual-drive hybrid model that combines physical information and data; preferably, the physical equation residual term includes the gas state equation residual, mass conservation residual, energy conservation residual, momentum conservation residual and heat conduction residual.
[0021] The compensation in the above step 3) includes dynamically adjusting the fusion weight of the dual-drive hybrid model combining physical information and data based on real-time environmental parameters, as well as implicit coupling of environmental variables with compressed air energy storage system parameters; preferably, the real-time environmental parameters are temperature, humidity or air pressure.
[0022] The specific implementation of step 4) above is:
[0023] 4.1) Simulation data training;
[0024] 4.2) High-value working condition training.
[0025] The specific implementation method of the above step 4.1) is: based on high-precision CFD simulation software simulation to generate full-operating condition simulation data, align the distribution differences between the simulation data and the real scene through the domain adversarial network, use the aligned simulation data to perform initial training on the dual-drive hybrid model that combines the compensated physical information and data, and improve the initial accuracy of the dual-drive hybrid model that combines the compensated physical information and data; the full-operating condition simulation data covers temperatures of -30°C to 50°C and humidity of 20% to 95%; preferably, the data volume of the full-operating condition simulation data is not less than 100,000 groups.
[0026] The specific implementation method of the above step 4.2) is: collect and screen high-value operating condition data to re-train the dual-drive hybrid model that combines the compensated physical information and data, so as to improve the final accuracy of the dual-drive hybrid model that combines the compensated physical information and data; the high-value operating condition data is temperature and humidity mutation point data and / or equipment extreme load data; preferably, the high-value operating condition data is not less than 2000 groups.
[0027] The method for constructing the compressed air energy storage system performance monitoring system further includes:
[0028] 5) Uploading the compressed air energy storage system performance monitoring system obtained in step 4) to the cloud to form a cloud-based compressed air energy storage system performance monitoring system.
[0029] The specific implementation of step 5) above is:
[0030] 5.1) Uploading the compressed air energy storage system performance monitoring system obtained in step 4) to the cloud;
[0031] 5.2) incrementally updating the data of the compressed air energy storage system performance monitoring system in the cloud; preferably, the specific implementation method of step 5.2) is to collect 24 hours of data cached at the edge, aggregate multi-node data features through federated learning, and upload the aggregated data to the cloud daily;
[0032] 5.3) Performing drift detection on the compressed air energy storage system performance monitoring system in the cloud. If the drift detection is qualified, a stable compressed air energy storage system performance monitoring system is formed; if the drift detection is unqualified, retraining the compressed air energy storage system performance monitoring system generated in step 4) and uploading it to the cloud; preferably, the drift detection indicator is the model prediction mean absolute error MAE or KL divergence.
[0033] A compressed air energy storage system performance monitoring system is constructed based on the method for constructing a compressed air energy storage system performance monitoring system as described above.
[0034] The advantages of the present invention are:
[0035] The present invention provides a method for constructing a performance monitoring system for a compressed air energy storage system and a monitoring system. The method includes: 1) constructing a dual-drive hybrid model based on a combination of physical information and data; 2) performing self-optimization learning on the dual-drive hybrid model combining physical information and data obtained in step 1) to obtain a self-optimized dual-drive hybrid model combining physical information and data; 3) compensating the prediction results of the self-optimized dual-drive hybrid model combining physical information and data obtained in step 2) based on environmental variables to obtain a compensated dual-drive hybrid model combining physical information and data; 4) training the compensated dual-drive hybrid model combining physical information and data obtained in step 3) using a data set to improve the output accuracy of the model and obtain a performance monitoring system for a compressed air energy storage system. Compared with traditional technologies that rely on a single physical or data model, the present invention significantly improves the accuracy and environmental adaptability of system state prediction by deeply integrating mechanisms and data-driven methods; at the same time, the present invention utilizes dynamic correction technology to effectively compensate for the influence of environmental variables, ensuring high accuracy while performing fast calculations, and providing reliable support for real-time decision-making at the edge. In addition, considering that traditional cloud-based centralized control is difficult to meet the needs of high-dynamic scenarios due to data transmission delays, the present invention achieves millisecond-level response speeds through localized closed-loop control on the edge. Traditional solutions are prone to loss of control when communication is interrupted, while the present invention relies on the autonomous decision-making capabilities of the edge to ensure stable operation of the system, while greatly reducing dependence on communication bandwidth and significantly improving system robustness and real-time performance. In response to the problem that traditional models are prone to performance degradation due to equipment aging or environmental changes and require frequent manual maintenance, the present invention uses self-learning and collaborative optimization technology to achieve online dynamic calibration of the model and maintain high-precision control capabilities for a long time. Obviously, the present invention is based on an efficient computing architecture on the edge, which greatly reduces energy consumption and operation and maintenance dependence, extends equipment life, and provides sustainable low-cost, highly reliable operation support for the system. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 This is a design idea diagram of the construction method of the compressed air energy storage system performance monitoring system provided by the present invention. DETAILED DESCRIPTION
[0037] This invention aims to build a monitoring and decision-making model for compressed air energy storage systems with environmental adaptation, multi-variable coupling optimization, and real-time control capabilities through the deep integration of physical models and deep learning, thereby solving the key bottlenecks of traditional technologies in terms of dynamic environmental adaptability, multi-physics field coupling modeling accuracy, and real-time control. Figure 1 Specifically, it includes: breaking through the lag of traditional static models and achieving millisecond-level real-time optimization response at the edge; dynamically balancing the joint optimal control of efficiency and equipment life through a physical data dual-drive architecture; using simulation-transfer learning collaboration and active learning mechanisms to solve cold start and small sample data dependency problems; establishing a cloud-edge collaborative incremental learning and drift detection mechanism to support long-term self-evolution in complex environments, and providing highly reliable and adaptive energy storage control solutions for new power systems.
[0038] The present invention provides a method for constructing a compressed air energy storage system performance monitoring system, the method comprising the following steps:
[0039] 1) Build a dual-drive hybrid model based on the combination of physical information and data. Specifically, the specific implementation method of step 1) is:
[0040] At the physical layer, a gas state equation is selected; preferably, the gas state equation is the ideal gas law or the van der Waals equation. At the same time, a basic thermodynamic differential equation is constructed based on the gas state equation, the first law of thermodynamics, and the second law of thermodynamics to ensure the physical rationality of the physical information data dual-driven hybrid model in data-sparse areas;
[0041] In terms of the data layer, the basic thermodynamic differential equations obtained in step 1.2) are embedded in the training process of the deep neural network in the form of physical constraints to learn high-dimensional complex mapping relationships that are difficult to cover by physical models, such as the nonlinear effects of environmental variables (such as humidity and temperature) on system performance (such as compressor wear and energy efficiency fluctuations), and finally a dual-driven hybrid model combining physical information and data is constructed; preferably, the deep neural network is LSTM or Transformer to adapt to the modeling of problems with time or state dependence characteristics; physical information is information obtained by calculating thermodynamic differential equations; data is supplementary information to the information calculated by thermodynamic equations, and the data includes simulation data, operating data of the compressed air energy storage system, data collected by sensors, and other experimental data.
[0042] The dual-driven hybrid model constructed by the present invention is based on the combination of physical information and data. By deeply embedding the mathematical structure of the thermodynamic equation into the computational graph of the neural network, it realizes the organic fusion of physical mechanism and data-driven. While retaining the flexibility of data-driven, it uses physical conservation laws to construct hard constraint boundaries, which is superior to the hybrid model in dynamic adaptability and extreme working condition extrapolation. Specifically, the model converts thermodynamic differential equations (such as the partial differential form of energy conservation) into topological constraints in the network structure: for example, embedding the mass conservation projection operator after the convolution layer, or constructing the weight distribution in the attention mechanism based on physical residuals. This design enables the neural network to simultaneously verify and maintain physical laws while performing data feature extraction. During the backpropagation process, the model not only optimizes the traditional loss function (such as data fitting error), but also introduces the Jacobian matrix based on the thermodynamic equation to correct the gradient direction, thereby ensuring that the update of the network parameters always remains within the physical feasible domain. Through this "equation as structure" strategy, the model not only has the expressive ability to handle complex working conditions (such as turbulent boundary layer separation), but also strictly abides by the basic laws of thermodynamics, achieving a dynamic balance between physical consistency and data generalization ability in input-output mapping.
[0043] A single physical or data model is built on the physical conservation laws of the compressed air energy storage system (mass / energy conservation, thermodynamic equations, etc.), and its prediction results have clear physical interpretability, which can ensure basic accuracy under known working conditions, but has poor predictability for unknown working conditions. The mechanism model provided by the present invention constructs a prediction benchmark based on the physical conservation laws to ensure the rationality of the basic logic; the data model captures the implicit nonlinear disturbances of the system (such as seal degradation, environmental mutation) (variable working conditions) in real time, and performs dynamic residual correction on the mechanism prediction. The two form a closed loop through two-way error compensation, which not only avoids the absurd output of the data model, but also corrects the idealized deviation of the mechanism model. In response to environmental variable disturbances, a dynamic weighting strategy of "steady-state mechanism priority, mutation data enhancement" is adopted to enable the system to actively perceive environmental changes and adaptively adjust the model coupling ratio to achieve negative feedback suppression of environmental disturbances and prediction errors.
[0044] 2) performing self-optimization learning on the dual-drive hybrid model combining the physical information and data obtained in step 1) to obtain a self-optimized dual-drive hybrid model combining the physical information and data. Specifically, the specific method of self-optimization learning in step 2) is to update the weights of the deep neural network and / or improve the output accuracy;
[0045] Among them, the specific implementation method of updating the deep neural network weights is:
[0046] Lightweight edge computing nodes are deployed on-site at the compressed air energy storage system to collect real-time operational data and environmental parameters. Using an online gradient descent algorithm, the edge nodes dynamically update the weights of the deep neural network in a dual-driven hybrid model that integrates physical information and data-driven learning, enabling adaptive optimization and continuous learning of the model. Preferably, operational data includes pressure, temperature, and power; environmental parameters include temperature, humidity, and atmospheric pressure.
[0047] The specific implementation method for improving output accuracy is: introducing physical equation residual terms in the loss function to improve the output accuracy of the dual-driven hybrid model that combines physical information and data, and forcing the output of the thermodynamic model to conform to the basic laws of thermodynamics; preferably, the physical equation residual terms include gas state equation residuals, mass conservation residuals, energy conservation residuals, momentum conservation residuals and heat conduction residuals.
[0048] 3) Based on the environmental variables, the prediction results of the dual-drive hybrid model combining the self-optimized physical information and data obtained in step 2) are compensated to obtain the compensated dual-drive hybrid model combining the physical information and data. Specifically, the compensation in step 3) includes dynamically adjusting the fusion weights of the dual-drive hybrid model combining the physical information and data based on the real-time environmental parameters, as well as implicit coupling between the environmental variables and the parameters of the compressed air energy storage system; preferably, the real-time environmental parameters are temperature, humidity or air pressure; exemplarily, the implicit coupling includes the material expansion effect caused by humidity changes, high temperature and humidity causing shortened compressor life or increased risk of electrical failure, large day and night temperature differences or rapid temperature fluctuations causing material thermal stress changes leading to seal aging or weld fatigue, wet air compression affecting compression efficiency, resulting in thermodynamic model deviations, etc. For example, dynamic coupling and quantitative analysis can be performed through multivariable coupling weight learning (automatically identifying key environmental variables (such as the synergistic effect of high temperature + high humidity) through the attention mechanism (Attention), and quantifying their impact weights on system efficiency and life) and life-efficiency joint optimization (designing a multi-objective reinforcement learning strategy to balance short-term energy efficiency and long-term equipment loss (for example, actively reducing power in high temperature environments to extend compressor life)), ultimately achieving compensation for the prediction results.
[0049] The dynamic correction adopted by the present invention is to make a residual correction to the initial value of the state prediction calculated by the mechanism model based on ideal assumptions (such as adiabatic conditions, no leakage, etc.), to compensate for the theoretical deviation caused by complex interference factors such as sealing loss and turbulent disturbance in actual operation. According to the real-time change amplitude of environmental variables (temperature / humidity / grid frequency), the weight distribution of the prediction results of the mechanism model and the data model is dynamically adjusted: when the environment is stable, the mechanism model is dominant (>70% weight), and when the environment changes suddenly, the weight of the data model is automatically increased to more than 50% to enhance the anti-interference ability. The compensation based on environmental variables adopted by the present invention: the dynamic disturbance of environmental variables will cause prediction deviations in traditional methods, especially when the environment fluctuates in extreme cases, it is difficult for a single model to meet the physical consistency and real-time anti-interference requirements at the same time. This invention uses dynamic correction technology to construct an environmental disturbance compensation mechanism based on real-time environmental parameters (temperature, humidity, air pressure, etc.) to dynamically adjust the fusion weights of the mechanism model and the data model. In a stable environment, the mechanism model is used to ensure the constraints of physical laws, and when the environment suddenly changes, the data model's ability to learn real-time disturbances is enhanced. At the same time, two-way parameter correction is performed for the implicit coupling between environmental variables and device parameters (such as the material expansion effect caused by humidity changes). This technology transforms environmental variables from "passive input" into "active compensation factors," forming a negative feedback relationship between the system prediction results and the environmental disturbance, ultimately achieving cross-scenario stability of environmental interference impact attenuation and prediction accuracy.
[0050] 4) The dual-drive hybrid model combining the compensated physical information and data obtained in step 3) is trained using the data set to improve the output accuracy of the model and obtain a compressed air energy storage system performance monitoring system. Specifically, this step is implemented as follows:
[0051] 4.1) Simulation data training; illustratively, during simulation data training, the following method can be used: generate full-operation simulation data based on high-precision CFD simulation software, align the distribution differences between the simulation data and the real scene through a domain adversarial network, and use the aligned simulation data to perform initial training on the dual-drive hybrid model that combines the compensated physical information and data, thereby improving the initial accuracy of the dual-drive hybrid model that combines the compensated physical information and data; the full-operation simulation data covers temperatures of -30°C to 50°C and humidity of 20% to 95%; preferably, the amount of full-operation simulation data is not less than 100,000 groups.
[0052] 4.2) High-value operating condition training. Exemplarily, the training method can be as follows: collect and screen high-value operating condition data to retrain the dual-drive hybrid model that combines compensated physical information and data, thereby improving the final accuracy of the dual-drive hybrid model that combines compensated physical information and data; the high-value operating condition data are temperature and humidity mutation point data and / or equipment extreme load data; preferably, the high-value operating condition data is not less than 2000 groups.
[0053] The method for constructing the compressed air energy storage system performance monitoring system provided by the present invention also includes:
[0054] 5) Uploading the compressed air energy storage system performance monitoring system obtained in step 4) to the cloud to form a cloud-based compressed air energy storage system performance monitoring system, specifically:
[0055] 5.1) Upload the compressed air energy storage system performance monitoring system obtained in step 4) to the cloud; illustratively, key summary data of the monitoring system (such as performance degradation trends and abnormal events) can be uploaded to the cloud, reducing bandwidth requirements by more than 90%.
[0056] 5.2) incrementally updating the data of the compressed air energy storage system performance monitoring system in the cloud; preferably, step 5.2) is implemented by collecting 24 hours of data cached at the edge, aggregating multi-node data features through federated learning (FedAvg), and uploading the aggregated data to the cloud regularly (exemplarily, daily);
[0057] 5.3) Drift detection is performed on the compressed air energy storage system performance monitoring system in the cloud. If the drift detection is qualified, a stable compressed air energy storage system performance monitoring system is formed; if the drift detection is unqualified, the compressed air energy storage system performance monitoring system generated in step 4) is retrained and uploaded to the cloud; preferably, the drift detection indicator is the model prediction mean absolute error MAE (Mean Absolute Error) or KL divergence (predicted wear rate vs. actual value). Exemplarily, drift detection can use the two indicators shown above. Both indicators are used to determine the error between the predicted value and the true value. If the error is greater than the threshold, an alarm mechanism will be triggered and the model on the edge will be retrained.
[0058] When in use, a lightweight version is generated through knowledge distillation and sent to the edge within 30 minutes, supporting zero-downtime updates. It should be noted that the edge computing of the present invention is the local end close to the source of data generation (such as sensors and device terminals).
[0059] While providing a method for constructing a compressed air energy storage system performance monitoring system as described above, the present invention also provides a compressed air energy storage system performance monitoring system constructed based on the construction method.
[0060] This invention breaks through the accuracy and speed bottlenecks of traditional single modeling methods by deeply integrating physical mechanisms and data-driven models, and achieves millisecond-level edge-end real-time decision-making in highly dynamic scenarios. Compared with existing technologies, its innovation lies in: building an environmentally adaptive dynamic multi-objective optimization mechanism to intelligently balance system efficiency, safety, and equipment life; adopting an edge-cloud collaborative architecture to improve control real-time performance and reliability while reducing communication dependence; introducing self-evolutionary learning technology to continuously calibrate model drift and ensure long-term operational stability. This solution significantly reduces operation and maintenance costs, extends the life cycle of equipment, and provides core technical support for large-scale energy storage applications in new power systems.
Claims
1. A method for constructing a compressed air energy storage system performance monitoring system, characterized by: The method for constructing the compressed air energy storage system performance monitoring system comprises the following steps: 1) Build a dual-driven hybrid model based on the combination of physical information and data; 2) performing self-optimization learning on the dual-drive hybrid model combining the physical information and data obtained in step 1) to obtain a self-optimized dual-drive hybrid model combining the physical information and data; 3) compensating the prediction result of the dual-drive hybrid model combining the self-optimized physical information and data obtained in step 2) based on the environmental variables to obtain a compensated dual-drive hybrid model combining the physical information and data; 4) The dual-drive hybrid model combining the compensated physical information and data obtained in step 3) is trained using the data set to improve the output accuracy of the model and obtain a compressed air energy storage system performance monitoring system.
2. The method for constructing a compressed air energy storage system performance monitoring system according to claim 1, characterized in that: The specific implementation of step 1) is: 1.1) Selecting a gas state equation; preferably, the gas state equation is the ideal gas law or the van der Waals equation; 1.2) Construct basic thermodynamic differential equations based on the gas state equation, the first law of thermodynamics, and the second law of thermodynamics; 1.3) Embedding the basic thermodynamic differential equations obtained in step 1.2) into the training process of a deep neural network in the form of physical constraints to construct a dual-driven hybrid model combining physical information and data; preferably, the deep neural network is an LSTM or a Transformer; the physical information is information calculated by the thermodynamic differential equations; the data is information supplementing the information calculated by the thermodynamic equations, and the data includes simulation data, operating data of the compressed air energy storage system, data collected by sensors, and other experimental data.
3. The method for constructing a compressed air energy storage system performance monitoring system according to claim 2, characterized in that: The specific method of self-optimization learning in step 2) is to update the weights of the deep neural network and / or improve the output accuracy; The specific implementation method of updating the weights of the deep neural network is: real-time collection of operating data and environmental parameters of the compressed air energy storage system, and dynamic updating of the weights of the deep neural network in the dual-drive hybrid model combining physical information and data through an online gradient descent algorithm; preferably, the operating data includes pressure, temperature and power; the environmental parameters include temperature, humidity and air pressure; The specific implementation method for improving the output accuracy is: introducing the physical equation residual term in the loss function to improve the output accuracy of the dual-drive hybrid model that combines physical information and data; preferably, the physical equation residual term includes the gas state equation residual, mass conservation residual, energy conservation residual, momentum conservation residual and heat conduction residual.
4. The method for constructing a compressed air energy storage system performance monitoring system according to claim 3, characterized in that: The compensation in step 3) includes dynamically adjusting the fusion weight of the dual-drive hybrid model combining physical information and data based on real-time environmental parameters, as well as implicit coupling of environmental variables with compressed air energy storage system parameters; preferably, the real-time environmental parameters are temperature, humidity or air pressure.
5. The method for constructing a compressed air energy storage system performance monitoring system according to claim 4, characterized in that: The specific implementation of step 4) is: 4.1) Simulation data training; 4.2) High-value working condition training.
6. The method for constructing a compressed air energy storage system performance monitoring system according to claim 5, characterized in that: The specific implementation method of step 4.1) is: generating full-condition simulation data based on high-precision CFD simulation software, aligning the distribution differences between the simulation data and the real scene through a domain adversarial network, and using the aligned simulation data to perform initial training on the dual-drive hybrid model that combines the compensated physical information and data, thereby improving the initial accuracy of the dual-drive hybrid model that combines the compensated physical information and data; The full-operating-condition simulation data covers data with a temperature range of -30°C to 50°C and a humidity range of 20% to 95%. Preferably, the data volume of the full-operating-condition simulation data is not less than 100,000 groups.
7. The method for constructing a compressed air energy storage system performance monitoring system according to claim 6, characterized in that: The specific implementation method of step 4.2) is: collecting and screening high-value operating condition data to re-train the dual-drive hybrid model that combines the compensated physical information and data, so as to improve the final accuracy of the dual-drive hybrid model that combines the compensated physical information and data; the high-value operating condition data is temperature and humidity mutation point data and / or equipment extreme load data; preferably, the high-value operating condition data is not less than 2000 groups.
8. The method for constructing a compressed air energy storage system performance monitoring system according to any one of claims 1 to 7, characterized in that: The method for constructing the compressed air energy storage system performance monitoring system further includes: 5) Uploading the compressed air energy storage system performance monitoring system obtained in step 4) to the cloud to form a cloud-based compressed air energy storage system performance monitoring system.
9. The method for constructing a compressed air energy storage system performance monitoring system according to claim 8, characterized in that: The specific implementation of step 5) is: 5.1) Uploading the compressed air energy storage system performance monitoring system obtained in step 4) to the cloud; 5.2) incrementally updating the data of the compressed air energy storage system performance monitoring system in the cloud; preferably, the specific implementation method of step 5.2) is to collect 24 hours of data cached at the edge, aggregate multi-node data features through federated learning, and upload the aggregated data to the cloud daily; 5.3) Performing drift detection on the compressed air energy storage system performance monitoring system in the cloud. If the drift detection is qualified, a stable compressed air energy storage system performance monitoring system is formed; if the drift detection is unqualified, retraining the compressed air energy storage system performance monitoring system generated in step 4) and uploading it to the cloud; preferably, the drift detection indicator is the model prediction mean absolute error MAE or KL divergence.
10. A compressed air energy storage system performance monitoring system constructed based on the method for constructing a compressed air energy storage system performance monitoring system according to claim 9.