Thermal management control method and system for energy storage power station based on deep neural network

Abstract

The present invention discloses a thermal management control method for an energy storage power station based on a deep neural network, comprising the steps of: obtaining a monitoring segment of the power output of the current energy storage power station, and using a time series prediction model based on the deep neural network to predict the power for the next time period, wherein the time series prediction model is obtained by training and learning an improved cross convolutional neural network model, wherein the improved cross convolutional neural network model comprises two parallel convolution branches, wherein the first branch uses a one-dimensional convolutional layer to extract local time series features of the power sequence, and the second branch uses a hole convolutional layer to expand the receptive field and capture the long-term trend of power changes; calculating the charge state and temperature state of the energy storage power station according to the predicted output power value of the energy storage power station in the next time period; using the coolant temperature, cabin temperature, ambient temperature and battery charge state of the current energy storage power station as inputs to an electric compressor speed setting model, and outputting the speed of the electric compressor.

Description

Technical Field

[0001] The present invention relates to the field of energy storage power stations, and in particular to a thermal management control method and system for energy storage power stations based on a deep neural network. Background Art

[0002] Existing thermal management technologies for energy storage power stations mostly use static threshold control or traditional PID algorithms, relying on historical experience data to adjust cooling system parameters. For example, some solutions directly adjust the compressor speed by monitoring the battery temperature, but do not consider the impact of future power fluctuations on the heat load, resulting in a delayed response. In addition, the following defects exist in the existing technology: 1. Insufficient power prediction accuracy: Traditional time series models are difficult to capture complex nonlinear relationships and cannot adapt to the high-frequency fluctuations in the dynamic power output of energy storage power stations; 2. Low efficiency in temperature state calculation: Temperature simulation calculations based on physical models are time-consuming and cannot be updated in real time; some data-driven methods such as linear regression have weak fusion capabilities for multi-source uncertainties (ambient temperature, coolant status); 3. Simplified control strategies: Relying on fixed rule bases or experience thresholds, lacking adaptive optimization capabilities, resulting in low energy efficiency. Summary of the invention

[0003] The main advantage of the present invention is that it provides a thermal management control method for an energy storage power station based on a deep neural network. Its time series prediction model for power prediction is based on an improved cross-convolutional neural network model, which can reduce prediction errors and help improve prediction accuracy.

[0004] Another advantage of the present invention is that it provides a thermal management control method for energy storage power stations based on deep neural networks. Its temperature state prediction model combines adaptive simulated annealing algorithm and DS evidence reasoning fusion for parameter optimization, which can reduce calculation time, improve the convergence speed of the temperature state prediction model, and improve the generalization ability of the temperature state prediction model.

[0005] Another advantage of the present invention is that it provides a thermal management control method for an energy storage power station based on a deep neural network, which uses a confidence rule base to construct an electric compressor speed setting model, combines expert knowledge with real-time data, and dynamically generates the speed of the electric compressor, thereby improving the adaptive ability and improving the system accuracy, which is beneficial to reducing the energy consumption of the compressor and improving the thermal management efficiency.

[0006] Accordingly, according to an embodiment of the present invention, a thermal management control method for an energy storage power station based on a deep neural network having at least one of the aforementioned advantages comprises the following steps:

[0007] S1. Obtain the monitoring fragment of the power output of the current energy storage power station, and use the time series prediction model based on deep neural network to predict the power for the next time period. The time series prediction model is obtained by training and learning with the improved cross convolutional neural network model. The improved cross convolutional neural network model includes two parallel convolution branches, the first branch uses a one-dimensional convolution layer to extract the local time series features of the power sequence, and the second branch uses a hole convolution layer to expand the receptive field and capture the long-term trend of power changes.

[0008] S2. Calculate the charge state and temperature state of the energy storage power station according to the predicted output power value of the energy storage power station in the next time period;

[0009] S3. The coolant temperature, cabin temperature, ambient temperature and battery charge state of the current energy storage power station are used as inputs of the electric compressor speed setting model, and the speed of the electric compressor is output.

[0010] In some embodiments of the present invention, step S1 comprises the steps of:

[0011] S11. Obtain historical monitoring data of power output of the energy storage power station;

[0012] S12, slice the time series of historical monitoring data using a sliding time window, and take the average value of the next k-th time window as the prediction target, so as to construct a time series prediction data set;

[0013] S13. Construct an improved cross-convolutional neural network model, including the following steps: construct a dual-channel parallel convolutional architecture, wherein the first branch uses a stacked one-dimensional convolutional layer, uses a 3×1 convolutional kernel to extract the local time series features of the power sequence to capture short-term charge and discharge fluctuations, and the second branch uses a hole convolutional layer to expand the receptive field, and configures 5×1 and 7×1 convolutional kernels respectively to capture the long-term trend of power changes; the dual-branch output is subjected to cross-channel feature fusion through a 1×1 convolutional kernel, connected to a multi-head self-attention module, and through a learnable query-key-value pair, the correlation between time segments is calculated and an attention weight matrix containing time series position encoding is generated to strengthen the feature weight of the power sudden change period; a gated recurrent unit is used to perform time series modeling on the weighted features and output the average power prediction value for the next time period through a fully connected layer;

[0014] S14. Use the constructed time series prediction dataset to train the model and adjust the model's hyperparameters according to the prediction results;

[0015] S15. When the set conditions for the end of training are met, the model training is ended and the model is deployed to the thermal management system of the energy storage power station.

[0016] In some embodiments of the present invention, the formula for calculating the state of charge of the energy storage power station is: , where Energy(t) is the remaining power of the energy storage station at time t, Power(t+1) is the predicted power output of the energy storage station at time t+1, △T is the length of the time segment, and Energy is the rated capacity of the energy storage station.

[0017] In some embodiments of the present invention, the temperature state of the energy storage power station is calculated by a temperature state prediction model, wherein the temperature state prediction model is obtained by training and learning a fuzzy DS evidence reasoning fusion model based on an adaptive simulated annealing algorithm.

[0018] In some embodiments of the present invention, in the process of training a temperature state prediction model, based on an adaptive simulated annealing algorithm and a fuzzy DS evidence reasoning fusion method, by setting initial parameters, an annealing mechanism is used to generate candidate solution combinations of penalty factors and variance parameters of radial basis kernel functions, and in each annealing stage, the local optimal solution of the temperature state prediction model is subjected to multi-source uncertainty fusion of DS evidence theory, the parameter search direction is dynamically adjusted through a probability distribution function, and iterative optimization is performed until the convergence conditions are met, and finally a penalty factor and variance parameter combination of the radial basis kernel function that minimizes the global error of the temperature state prediction model is obtained.

[0019] In some embodiments of the present invention, the training samples of the temperature state prediction model are obtained through thermal management finite element simulation calculation.

[0020] In some embodiments of the present invention, a confidence rule base is used to construct a speed setting model for an electric compressor, and the speed of the electric compressor is dynamically generated by combining expert knowledge with real-time data.

[0021] Accordingly, the present invention further provides a thermal management control system for an energy storage power station based on a deep neural network, which is used to implement the thermal management control method for an energy storage power station based on a deep neural network, comprising:

[0022] The energy storage power station power prediction module is used to obtain the monitoring fragment of the current energy storage power station power output and use the time series prediction model based on the improved cross convolutional neural network to predict the power for the next time period;

[0023] The energy storage power station state calculation module calculates the charge state and temperature state of the energy storage power station according to the predicted output power value of the energy storage power station in the next time period; and

[0024] The electric compressor speed setting module integrates the coolant temperature, cabin temperature, ambient temperature and battery charge state of the current energy storage power station based on the confidence rule base, and dynamically outputs the speed of the electric compressor through expert knowledge and rule reasoning.

[0025] Accordingly, the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the thermal management control method of an energy storage power station based on a deep neural network is implemented.

[0026] Accordingly, the present invention further provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the thermal management control method of the energy storage power station based on a deep neural network when executing the computer program.

[0027] The above and other advantages of the present invention will be fully reflected in conjunction with the following description and the accompanying drawings.

[0028] The above and other advantages and features of the present invention are fully reflected in the following detailed description of the present invention and the accompanying drawings.

[0029] The content of the invention is not to be regarded as identifying the essential technical features of the invention, nor is it to be regarded as limiting the protection scope of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a schematic diagram of a thermal management control method for an energy storage power station based on a deep neural network according to an embodiment of the present invention.

[0031] Figure 2 is a schematic diagram of step S1 of a thermal management control method for an energy storage power station based on a deep neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following description is provided to enable those skilled in the art to implement the present invention. Other obvious replacements, modifications and variations may occur to those skilled in the art. Therefore, the scope of protection of the present invention should not be limited to the exemplary embodiments described herein.

[0033] Those skilled in the art should understand that, unless otherwise specified herein, the terms "one" and "an" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the element may be multiple.

[0034] Those skilled in the art should understand that, unless otherwise specified herein, the directions or positions referred to by the terms "longitudinal", "lateral", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. are based on the directions or positions shown in the drawings, and are only for the convenience of describing the present invention, and do not indicate or imply that the devices or elements involved must have a specific direction or position. Therefore, the above terms should not be understood as limiting the present invention.

[0035] Refer to the attached specification of the present invention Figure 1 and attached Figure 2 , a thermal management control method for an energy storage power station based on a deep neural network according to an embodiment of the present invention is explained. The thermal management control method for an energy storage power station based on a deep neural network comprises the following steps:

[0036] S1. Obtain the monitoring fragment of the power output of the current energy storage power station, and use the time series prediction model based on deep neural network to predict the power for the next time period. The time series prediction model is obtained by training and learning with the improved cross convolutional neural network model. The improved cross convolutional neural network model includes two parallel convolution branches, the first branch uses a one-dimensional convolution layer to extract the local time series features of the power sequence, and the second branch uses a hole convolution layer to expand the receptive field and capture the long-term trend of power changes.

[0037] S2. Calculate the charge state and temperature state of the energy storage power station according to the predicted output power value of the energy storage power station in the next time period;

[0038] S3. The coolant temperature, cabin temperature, ambient temperature and battery charge state of the current energy storage power station are used as inputs of the electric compressor speed setting model, and the speed of the electric compressor is output.

[0039] It can be understood that since the time series prediction model is obtained by training and learning with the improved cross convolutional neural network model, and the improved cross convolutional neural network model includes two parallel convolution branches, the first branch uses a one-dimensional convolution layer to extract the local time series features of the power sequence, and the second branch uses a hole convolution layer to expand the receptive field and capture the long-term trend of power changes. In this way, the prediction error can be reduced, which is conducive to improving the prediction accuracy.

[0040] Furthermore, the time series prediction model used for power prediction of energy storage power stations collects historical power data in real time, uses a sliding time window to cut continuous data into segments, and predicts the average power value of the next time period in the future. The core design of the time series prediction model includes two parallel convolution branches, where the first branch uses a one-dimensional convolution layer to extract the local time series features of the power sequence (such as short-term charging and discharging fluctuations), and the second branch uses a hole convolution layer to expand the receptive field and capture the long-term trend of power changes (such as continuous high-load operation mode). After the output features of the two branches are fused, they are input into the multi-head self-attention module to dynamically assign importance weights to different time segments and highlight the impact of key periods such as power surges or drops.

[0041] Specifically, as attached Figure 2 As shown, step S1 includes the steps of:

[0042] S11. Obtain historical monitoring data of power output of the energy storage power station;

[0043] S12, slice the time series of historical monitoring data using a sliding time window, and take the average value of the next k-th time window as the prediction target, so as to construct a time series prediction data set;

[0044] S13. Construct an improved cross-convolutional neural network model, including the following steps: construct a dual-channel parallel convolutional architecture, wherein the first branch uses a stacked one-dimensional convolutional layer, uses a 3×1 convolutional kernel to extract the local time series features of the power sequence to capture short-term charge and discharge fluctuations, and the second branch uses a hole convolutional layer to expand the receptive field, and configures 5×1 and 7×1 convolutional kernels respectively to capture the long-term trend of power changes; the dual-branch output is subjected to cross-channel feature fusion through a 1×1 convolutional kernel, connected to a multi-head self-attention module, and through a learnable query-key-value pair, the correlation between time segments is calculated and an attention weight matrix containing time series position encoding is generated to strengthen the feature weight of the power sudden change period; a gated recurrent unit is used to perform time series modeling on the weighted features and output the average power prediction value for the next time period through a fully connected layer;

[0045] S14. Use the constructed time series prediction dataset to train the model and adjust the model's hyperparameters according to the prediction results;

[0046] S15. When the set conditions for the end of training are met, the model training is ended and the model is deployed to the thermal management system of the energy storage power station.

[0047] Furthermore, in some embodiments of the present invention, the formula for calculating the state of charge of the energy storage power station is: , where Energy(t) is the remaining power of the energy storage station at time t, Power(t+1) is the predicted power output of the energy storage station at time t+1, △T is the length of the time segment, and Energy is the rated capacity of the energy storage station.

[0048] Preferably, the present invention calculates the temperature state of the energy storage power station through a temperature state prediction model, wherein the temperature state prediction model is obtained by training and learning using a fuzzy DS evidence reasoning fusion model based on an adaptive simulated annealing algorithm. It can be understood that the temperature state prediction model combines the adaptive simulated annealing algorithm and the DS evidence reasoning fusion to perform parameter optimization, which can reduce the computational time, improve the convergence speed of the temperature state prediction model, and improve the generalization ability of the temperature state prediction model.

[0049] Furthermore, the training data of the temperature state prediction model is generated by finite element simulation to simulate the battery temperature field distribution under different ambient temperatures, coolant flow rates and charge and discharge rates. The adaptive simulated annealing algorithm is used to optimize the parameters of the temperature state prediction model. The algorithm gradually cools down from the high temperature initial state, randomly generates candidate parameter combinations, and uses the prediction error as an evaluation indicator. Unlike traditional annealing, the present invention introduces fuzzy DS evidence theory in each iteration, quantifies the influence weights of multi-source uncertainties such as simulation errors and sensor noise, and dynamically adjusts the parameter search direction through evidence fusion (for example, prioritizes the optimization of kernel function parameters that are sensitive to coolant temperature). If optimization stagnation is detected, the temperature reset mechanism is triggered to expand the search range to avoid local optimality. The present invention combines the global search capability of the annealing algorithm with the uncertainty fusion of the evidence theory to reduce the time consumption of temperature calculation and improve the generalization ability of the model.

[0050] Specifically, in some embodiments of the present invention, in the process of training the temperature state prediction model, based on the adaptive simulated annealing algorithm and the fuzzy DS evidence reasoning fusion method, by setting the initial parameters, the annealing mechanism is used to generate the candidate solution combination of the penalty factor and the variance parameter of the radial basis kernel function, and the multi-source uncertainty fusion of the DS evidence theory is performed on the local optimal solution of the temperature state prediction model in each annealing stage, and the parameter search direction is dynamically adjusted through the probability distribution function, and iterative optimization is performed until the convergence condition is met, and finally the penalty factor and the variance parameter combination of the radial basis kernel function that minimizes the global error of the temperature state prediction model are obtained. Preferably, the training samples of the temperature state prediction model are obtained through thermal management finite element simulation operations.

[0051] In order to overcome the adaptive defects of static threshold control, the present invention constructs a fuzzy rule base that integrates expert knowledge and real-time data. Specifically, the present invention uses a confidence rule base to construct an electric compressor speed setting model, combines expert knowledge with real-time data, and dynamically generates the speed of the electric compressor, thereby improving the adaptive ability and the accuracy of the system, which is conducive to reducing the energy consumption of the compressor and improving the thermal management efficiency.

[0052] According to another aspect of the present invention, the present invention further provides a thermal management control system for an energy storage power station based on a deep neural network, which is used to implement the thermal management control method for an energy storage power station based on a deep neural network, comprising:

[0053] The energy storage power station power prediction module is used to obtain the monitoring fragment of the current energy storage power station power output and use the time series prediction model based on the improved cross convolutional neural network to predict the power for the next time period;

[0054] The energy storage power station state calculation module calculates the charge state and temperature state of the energy storage power station according to the predicted output power value of the energy storage power station in the next time period; and

[0055] The electric compressor speed setting module integrates the coolant temperature, cabin temperature, ambient temperature and battery charge state of the current energy storage power station based on the confidence rule base, and dynamically outputs the speed of the electric compressor through expert knowledge and rule reasoning.

[0056] According to another aspect of the present invention, the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the thermal management control method of an energy storage power station based on a deep neural network is implemented.

[0057] According to another aspect of the present invention, the present invention further provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the thermal management control method for an energy storage power station based on a deep neural network is implemented.

[0058] Those skilled in the art will appreciate that the above embodiments are merely examples, wherein features of different embodiments may be combined with each other to obtain implementation methods that are easily conceivable based on the contents disclosed in the present invention but are not explicitly indicated in the drawings.

[0059] Those skilled in the art should understand that the above description and the embodiments shown in the drawings are only for illustrative explanation of the present invention, rather than for limitation of the present invention. All equivalent implementations, modifications and improvements within the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A thermal management control method for energy storage power station based on deep neural network, characterized in that: Includes steps: S1. Obtain the monitoring fragment of the power output of the current energy storage power station, and use the time series prediction model based on deep neural network to predict the power for the next time period. The time series prediction model is obtained by training and learning with the improved cross convolutional neural network model. The improved cross convolutional neural network model includes two parallel convolution branches, the first branch uses a one-dimensional convolution layer to extract the local time series features of the power sequence, and the second branch uses a hole convolution layer to expand the receptive field and capture the long-term trend of power changes. S2. Calculate the state of charge and temperature of the energy storage power station according to the predicted output power value of the energy storage power station in the next time period, wherein the temperature state of the energy storage power station is calculated by a temperature state prediction model, wherein the temperature state prediction model is obtained by training and learning a fuzzy DS evidence reasoning fusion model based on an adaptive simulated annealing algorithm; S3. The coolant temperature, cabin temperature, ambient temperature and battery charge state of the current energy storage power station are used as inputs of the electric compressor speed setting model, and the speed of the electric compressor is output. The electric compressor speed setting model is constructed using a confidence rule base, and the speed of the electric compressor is dynamically generated by combining expert knowledge and real-time data.

2. The thermal management control method for energy storage power station based on deep neural network according to claim 1 is characterized in that: Step S1 includes the steps of: S11. Obtain historical monitoring data of power output of the energy storage power station; S12, slice the time series of historical monitoring data using a sliding time window, and take the average value of the next k-th time window as the prediction target, so as to construct a time series prediction data set; S13. Construct an improved cross convolutional neural network model, including the steps of: constructing a dual-channel parallel convolution architecture, wherein the first branch uses a stacked one-dimensional convolution layer, uses a 3×1 convolution kernel to extract the local timing features of the power sequence to capture short-term charge and discharge fluctuations, and the second branch uses a hole convolution layer to expand the receptive field, and configures 5×1 and 7×1 convolution kernels respectively to capture the long-term trend of power changes; The dual-branch output is fused with cross-channel features through a 1×1 convolution kernel and connected to a multi-head self-attention module. Through learnable query-key-value pairs, the correlation between time segments is calculated and an attention weight matrix containing temporal position encoding is generated to strengthen the feature weight of the power sudden change period. The gated recurrent unit is used to perform time series modeling on the weighted features and the average power prediction value for the next time period is output through the fully connected layer; S14. Use the constructed time series prediction dataset to train the model and adjust the model's hyperparameters according to the prediction results; S15. When the set conditions for the end of training are met, the model training is ended and the model is deployed to the thermal management system of the energy storage power station.

3. The thermal management control method for energy storage power station based on deep neural network according to claim 1 is characterized in that: The formula for calculating the state of charge of the energy storage power station is: , where Energy(t) is the remaining power of the energy storage station at time t, Power(t+1) is the predicted power output of the energy storage station at time t+1, △T is the length of the time segment, and Energy is the rated capacity of the energy storage station.

4. The thermal management control method for energy storage power station based on deep neural network according to claim 1 is characterized in that: In the process of training the temperature state prediction model, based on the adaptive simulated annealing algorithm and the fuzzy DS evidence reasoning fusion method, by setting the initial parameters, the annealing mechanism is used to generate candidate solution combinations of the penalty factor and the variance parameter of the radial basis kernel function. In each annealing stage, the multi-source uncertainty fusion of the DS evidence theory is performed on the local optimal solution of the temperature state prediction model. The parameter search direction is dynamically adjusted through the probability distribution function. The optimization is iteratively optimized until the convergence conditions are met. Finally, the penalty factor and the variance parameter combination of the radial basis kernel function that minimizes the global error of the temperature state prediction model is obtained.

5. The thermal management control method for energy storage power station based on deep neural network according to claim 4 is characterized in that: The training samples of the temperature state prediction model are obtained through thermal management finite element simulation.

6. A thermal management control system for an energy storage power station based on a deep neural network, used to implement the thermal management control method for an energy storage power station based on a deep neural network as described in any one of claims 1 to 5, characterized in that: include: The energy storage power station power prediction module is used to obtain the monitoring fragment of the current energy storage power station power output and use the time series prediction model based on the improved cross convolutional neural network to predict the power for the next time period; The energy storage power station state calculation module calculates the charge state and temperature state of the energy storage power station according to the predicted output power value of the energy storage power station in the next time period; and The electric compressor speed setting module integrates the coolant temperature, cabin temperature, ambient temperature and battery charge state of the current energy storage power station based on the confidence rule base, and dynamically outputs the speed of the electric compressor through expert knowledge and rule reasoning.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the thermal management control method of an energy storage power station based on a deep neural network as described in any one of claims 1 to 5 is implemented.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the thermal management control method for an energy storage power station based on a deep neural network as described in any one of claims 1 to 5 is implemented.

Images