Energy storage EMS system SOC balance control system and method based on artificial intelligence

Abstract

The invention relates to the technical field of energy storage EMS, in particular to an energy storage EMS system SOC balance control system and method based on artificial intelligence, and the system comprises a monitoring system which is used for collecting the voltage, current, temperature and SOC parameters of a battery unit; the balance control system is used for carrying out balance control on the SOC of each battery unit in the battery pack through a reinforcement learning algorithm RStudy; the real-time regulation and control system is used for realizing real-time regulation of an SOC balance control strategy through a random forest algorithm; and the fault diagnosis and prevention system performs fault diagnosis through a random forest algorithm and performs prevention by adjusting a charging and discharging strategy. According to the design of the invention, a reinforcement learning algorithm RStudy is adopted to realize dynamic adjustment of the SOC of the battery unit, and the reinforcement learning algorithm RStudy can automatically learn an optimal decision strategy through interaction with the environment; the real-time reward and the long-term target can be balanced, excessive discharging or charging is avoided, battery aging is reduced, the optimal decision can be made in an uncertain environment, and the robustness and adaptability of the system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage EMS, and in particular to an energy storage EMS system (electric energy management system) SOC (battery state of charge) balancing control system and method based on artificial intelligence. Background Art

[0002] A battery pack is composed of multiple battery cells. The initial performance of each cell may be similar, but during use, due to differences in manufacturing processes, different aging rates, operating temperature and other factors, the state of charge (SOC) between cells will gradually differ. If the EMS system without SOC balancing control cannot actively correct these differences, the battery cells with poor performance will become a bottleneck, thereby affecting the energy storage efficiency of the overall system; and an unbalanced SOC will cause premature aging of the battery cells. Battery cells that often operate at a high SOC state will experience more severe capacity decay, while cells in a low SOC state may suffer damage from deep discharge, ultimately shortening the service life of the entire battery pack. Summary of the invention

[0003] The purpose of the present invention is to provide an energy storage EMS system SOC balancing control system and method based on artificial intelligence to solve the problem of battery overcharging or discharging in the above-mentioned background technology, so as to reduce the performance degradation and failure rate of the battery pack.

[0004] The energy storage EMS system SOC balancing control system based on artificial intelligence of the present invention comprises

[0005] A monitoring system that collects voltage, current, temperature, and SOC parameters of the battery cells;

[0006] The balanced control system uses the reinforcement learning algorithm RStudy to balance the SOC of each battery cell in the battery pack, and then introduces the temperature data of the battery cell into the mean square error loss function optimized by the battery temperature weight. According to the temperature condition corresponding to the sample, the contribution of the sample to the total loss function is adjusted. If the prediction at a certain temperature is very important, then w i It will be set higher, so as to give a larger penalty to the prediction error of the sample when calculating the total loss, so as to achieve the accuracy of predicting the SOC of the battery cell under different temperature conditions;

[0007] Among them, the optimized mean square error loss function is:

[0008]

[0009] Among them, w represents the weight, represents the weighted loss function, Y represents the actual SOC value, Represents the predicted SOC value, Y i represents the actual SOC value of the i-th sample, represents the predicted SOC value of the i-th sample, w i represents the temperature weight of the i-th sample;

[0010] A real-time control system, wherein the real-time control system realizes real-time adjustment of the SOC balance control strategy through a random forest algorithm; a fault diagnosis and prevention system, wherein the fault diagnosis and prevention system performs fault diagnosis through a random forest algorithm and takes preventive measures by adjusting the charge and discharge strategy.

[0011] As a further improvement of the technical solution, the balanced control system (1) comprises a data prediction module, a dynamic adjustment module and a target optimization module;

[0012] The data prediction module establishes the SOC prediction model of the battery cell through the long short-term memory network LSTM, analyzes the historical charge and discharge data of the battery cell and uses it as a reference to train the SOC prediction model. The trained SOC prediction model is then integrated into the real-time monitoring system to predict the SOC change trend of the battery cell by comparing the real-time monitoring data with the historical charge and discharge data.

[0013] The dynamic adjustment module uses the reinforcement learning algorithm RStudy to dynamically adjust the actual SOC and predicted SOC of the battery cell, and realizes balanced control of SOC by adjusting the charge and discharge power of the battery cell, so that the actual SOC is as close as possible to the predicted SOC, thus achieving balanced control;

[0014] The target optimization module finds the optimal SOC balance control strategy through genetic algorithm.

[0015] As a further improvement of the technical solution, the data prediction module establishes a SOC prediction model of the battery cell through a long short-term memory network LSTM, involving the following steps:

[0016] S3.1. Collect the historical charge and discharge data of the battery cell and the corresponding SOC data to form a data set, which includes the voltage, current, and temperature parameters of the battery, and preprocess the data set to divide the data set into a training set, a validation set, and a test set. The training set is a data set used to train the LSTM model. The data set includes the voltage, current, temperature parameters of the battery, and the state of charge SOC of the battery cell. The LSTM model learns the mapping relationship between input and output through the training set. In this process, the LSTM model will try to minimize the mean square error loss function on the training set to learn how to make correct predictions. The validation set is used to adjust the hyperparameters of the LSTM model, which include the number of hidden layers in the neural network, the learning rate, etc., and to prevent overfitting. The test set is used to finally evaluate the generalization ability of the LSTM model. This is an independent evaluation of the model performance and does not participate in any training or hyperparameter tuning process. The results of the test set give the performance level that the LSTM model may achieve in practical applications.

[0017] S3.2. Use the deep learning framework TensorFlow to define the LSTM model architecture, and optimize the mean square error loss function by introducing battery temperature weights to achieve accurate prediction under different temperature conditions;

[0018] S3.3, train the model using the training set data;

[0019] S3.4. Integrate the trained model into the real-time monitoring system and predict the SOC change trend based on the real-time monitoring data.

[0020] As a further improvement of the technical solution, the dynamic adjustment module adopts the reinforcement learning algorithm RStudy to realize the dynamic adjustment of the battery cell SOC, involving the following steps:

[0021] S4.1. Define a state space including the current SOC, predicted SOC, temperature, and charge and discharge state of the battery cell. The state set of each battery cell constitutes the state space of the overall system; define an action space for adjusting the charge and discharge power of the battery cell;

[0022] S4.2. Design a reward function that encourages energy balance between battery cells. When the SOC of a battery cell is close to the target value, a positive reward is given, otherwise a negative reward is given.

[0023] S4.3, initialize the Q table, assign an initial Q value to each state-action combination in the state space and action space, that is, the sum of future rewards expected from taking a certain action in a certain state, and the Q value is a constant value;

[0024] S4.4. After each iteration, the action is executed to obtain an immediate reward and a new state, the Q value is updated according to the Bellman equation, and the update of the constraint parameter optimization equation is introduced, that is, the action that maximizes the Q value is selected as the next action; S4.5. The Softmax strategy is used to balance exploration and utilization, which allows the algorithm to find a balance between trying new actions and repeating actions with known high Q values, that is, the action with the highest Q value is selected with a certain probability;

[0025] S4.6. When the change in Q value becomes very small, that is, the algorithm no longer significantly updates the Q table, it can be considered that the learning process has converged. The strategy at this time is considered to be optimized and can effectively control the SOC of the battery cell and achieve balanced control.

[0026] As a further improvement of the present technical solution, in S4.4, the optimized Bellman equation updates the Q value, where the learning rate determines the speed at which the algorithm updates the Q value. If the learning rate is set too high, the algorithm may jump too much when updating the Q value, resulting in an unstable learning process or even divergence. By introducing a limit parameter to optimize the update of the equation, limiting the size of the learning rate ensures that each update is not too aggressive, which helps the algorithm converge to the optimal strategy and helps the algorithm generalize better during the training process, specifically:

[0027] Q(s,a)←Q(s,a)+Pα[R+γmaxa′Q(s′,a′)-Q(s,a)];

[0028] Where Q(s, a) is the expected long-term reward of taking action a to adjust the charge and discharge power of the battery cell in state space s, R is the immediate reward obtained immediately after taking the action, P is the restriction parameter, α is the learning rate, γ is the discount factor, s′ is the new state after taking the action, and a′ is the best action among all actions in the new state.

[0029] As a further improvement of the technical solution, the target optimization module considers minimizing SOC difference, maximizing battery life, and optimizing energy utilization efficiency through a genetic algorithm to find the optimal balancing strategy, involving the following steps:

[0030] S6.1. Set up a multi-objective optimization framework, with the objectives including minimizing the SOC difference between battery cells, maximizing the battery pack life, and optimizing energy utilization efficiency;

[0031] S6.2, define multi-objective function using weighted sum method;

[0032] S6.3, select a genetic algorithm to solve the objective function;

[0033] S6.4. Evaluate the solution set generated by the algorithm and select the optimal solution that meets the multi-objective optimization criteria.

[0034] As a further improvement of the technical solution, in S6.2, the weighted summation method is specifically:

[0035] F(x)=w1f1(x)+w2f2(x)+w3f3(x)

[0036] Among them, F(x) represents the weighted sum value, f1(x) represents minimizing the SOC difference between battery cells, f2(x) represents maximizing the battery pack life, f3(x) represents optimizing the energy utilization efficiency, and w1, w2, and w3 represent the target weights.

[0037] As a further improvement of the technical solution, in S6.3, the genetic algorithm is used to solve the function specifically as follows:

[0038] S6.31, generate W solutions as the initial population, each solution represents a SOC equilibrium strategy;

[0039] S6.32, calculating the fitness value of each solution by weighted sum method;

[0040] S6.33, using tournament selection, select individuals from the current population to enter the breeding pool, and the probability of selection is proportional to the fitness of the individual;

[0041] S6.34, Generate new individuals through crossover and mutation operations. The crossover operation simulates the genetic process of organisms, while the mutation operation introduces new genetic mutations.

[0042] S6.35, replace old individuals with newly generated individuals to form a new population;

[0043] S6.36. If the quality of the solution reaches the predetermined standard A, analyze the solutions in the final population and select the optimal solution as the SOC equilibrium strategy.

[0044] As a further improvement of the technical solution, the real-time control system performs real-time adjustment and control of the equilibrium control strategy through a random forest algorithm, involving the following steps:

[0045] S9.1. Collecting operating data of the battery unit, the operating data including voltage, current, temperature and SOC parameters;

[0046] S9.2. Preprocess the data and use the collected data to train the random forest model;

[0047] S9.3. Select the parameters of the random forest, including the number of trees m, the maximum depth of the tree w, the minimum number of samples s required for node splitting, and perform cross-validation;

[0048] S9.4. Collect real-time battery status data and input the data into the trained random forest model for prediction;

[0049] S9.5. Adjust the balance control strategy in real time according to the prediction results of the model.

[0050] As a further improvement of the technical solution, in S9.4, the specific algorithm for inputting the real-time battery status data into the random forest model is:

[0051] S9.41. Use sample x to make predictions and get numerical predictions v b ;

[0052] S9.42. Calculate the average of the predicted values ​​to obtain the final prediction result:

[0053]

[0054] Among them, x represents the input real-time battery status data, v b It represents the SOC numerical result obtained after the input sample x is predicted, y represents the final prediction result of the input sample x, b represents an integer from 1 to B, and B represents the total number of predictions.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] 1. In the SOC balancing control system of the energy storage EMS system based on artificial intelligence, the reinforcement learning algorithm RStudy is used to realize the dynamic adjustment of the battery cell SOC. The reinforcement learning algorithm RStudy can automatically learn the optimal decision-making strategy through interaction with the environment; it can balance immediate rewards and long-term goals, avoid excessive discharge or charging, reduce battery aging, and make the best decisions in an uncertain environment, thereby improving the robustness and adaptability of the system.

[0057] 2. In the SOC balancing control system of the energy storage EMS system based on artificial intelligence, the random forest algorithm is used to accurately predict the SOC state and voltage, current, and temperature of the battery cell, providing a basis for dynamically adjusting the control strategy and reducing the risk of overfitting through a diverse set of decision trees; the prediction model is quickly adjusted according to real-time data, quickly adapting to changes in battery status, and timely adjusting the balancing strategy to maintain the health of the battery pack and extend its service life. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is the overall flow chart of the present invention;

[0059] The meaning of each number in the figure is:

[0060] 1. Balance control system; 2. Real-time control system; 3. Fault diagnosis and prevention system; 4. Monitoring system. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments 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 creative work are within the scope of protection of the present invention.

[0062] Example

[0063] See also Figure 1 As shown, an energy storage EMS system SOC balancing control system based on artificial intelligence is provided, including a monitoring system 4, wherein the monitoring system collects voltage, current, temperature and SOC parameters of the battery unit;

[0064] Among them, the SOC parameter represents the state of charge of the battery cell;

[0065] The balancing control system 1 uses the reinforcement learning algorithm RStudy to balance the SOC of each battery cell in the battery pack, and then introduces the temperature data of the battery cell into the mean square error loss function of the battery temperature weight optimization. According to the temperature condition corresponding to the sample, the contribution of the sample to the total loss function is adjusted. If the prediction at a certain temperature is very important, then w i It will be set higher, so as to give a larger penalty to the prediction error of the sample when calculating the total loss, so as to achieve the accuracy of predicting the SOC of the battery cell under different temperature conditions;

[0066] Among them, the optimized mean square error loss function is:

[0067]

[0068] Among them, w represents the weight, represents the weighted loss function, Y represents the actual SOC value, Represents the predicted SOC value, Y i represents the actual SOC value of the i-th sample, represents the predicted SOC value of the i-th sample, w i represents the temperature weight of the i-th sample;

[0069] A real-time control system 2, wherein the real-time control system 2 realizes real-time adjustment of the SOC balance control strategy through a random forest algorithm; a fault diagnosis and prevention system 3, wherein the fault diagnosis and prevention system 3 performs fault diagnosis through a random forest algorithm and takes preventive measures by adjusting the charge and discharge strategy.

[0070] In this embodiment, the balanced control system 1 includes a data prediction module, a dynamic adjustment module and a target optimization module;

[0071] The data prediction module establishes the SOC prediction model of the battery cell through the long short-term memory network LSTM, analyzes the historical charge and discharge data of the battery cell and uses it as a reference to train the SOC prediction model. The trained SOC prediction model is then integrated into the real-time monitoring system to predict the SOC change trend of the battery cell by comparing the real-time monitoring data with the historical charge and discharge data.

[0072] The dynamic adjustment module uses the reinforcement learning algorithm RStudy to dynamically adjust the actual SOC and predicted SOC of the battery cell, and realizes balanced control of SOC by adjusting the charge and discharge power of the battery cell, so that the actual SOC is as close as possible to the predicted SOC, thus achieving balanced control;

[0073] The target optimization module finds the optimal SOC balance control strategy through genetic algorithm;

[0074] Among them, the data prediction module establishes the SOC prediction model of the battery cell through the long short-term memory network LSTM. LSTM can help the network selectively remember or forget information, so as to effectively process the long-term dependency in the sequence data and reduce the influence of the battery's current SOC on the past charging and discharging cycles; it can capture and model the complex nonlinear relationship between the battery's SOC and other parameters (such as temperature, current, voltage, etc.). LSTM can adaptively adjust its internal state according to the input data to more accurately predict the future SOC value. The data prediction module establishes the SOC prediction model of the battery cell through the long short-term memory network LSTM, involving the following steps:

[0075] S3.1. Collect historical charge and discharge data of the battery unit and the corresponding SOC data to form a data set, which includes the voltage, current, and temperature parameters of the battery, and pre-process the data set to divide the data set into a training set, a validation set, and a test set;

[0076] The training set is a data set used to train the LSTM model. The data set includes the battery voltage, current, temperature parameters, and the state of charge (SOC) of the battery cell. The LSTM model learns the mapping relationship between input and output through the training set. In this process, the LSTM model will try to minimize the mean square error loss function on the training set to learn how to make correct predictions. The validation set is used to adjust the hyperparameters of the LSTM model, which include the number of hidden layers in the neural network, the learning rate, etc., and to prevent overfitting. The test set is used to finally evaluate the generalization ability of the LSTM model. This is an independent evaluation of the model performance and does not participate in any training or hyperparameter tuning process. The results of the test set give the performance level that the LSTM model may achieve in practical applications.

[0077] S3.2. Use the deep learning framework TensorFlow to define the LSTM model architecture, and optimize the mean square error loss function by introducing battery temperature weights to achieve accurate prediction of SOC under different temperature conditions;

[0078] S3.3, train the model using the training set data;

[0079] S3.4, integrate the trained model into the real-time monitoring system and predict the SOC change trend based on the real-time monitoring data;

[0080] Among them, the dynamic adjustment module uses the reinforcement learning algorithm RStudy to realize the dynamic adjustment of the battery cell SOC. The reinforcement learning algorithm RStudy can automatically learn the optimal decision-making strategy through interaction with the environment; it can dynamically adjust the SOC target based on the current state of the battery (such as SOC, temperature, charge and discharge rate, etc.) and external conditions (such as grid price, user demand, etc.) to optimize battery life, cost efficiency or energy utilization efficiency, involving the following steps:

[0081] S4.1. Define a state space including the current SOC, predicted SOC, temperature, and charge and discharge state of the battery cell. The state set of each battery cell constitutes the state space of the overall system; define an action space for adjusting the charge and discharge power of the battery cell;

[0082] S4.2. Design a reward function that encourages energy balance between battery cells. When the SOC of a battery cell is close to the target value, a positive reward is given, otherwise a negative reward is given.

[0083] S4.3, initialize the Q table, assign an initial Q value to each state-action combination in the state space and action space, that is, the sum of future rewards expected from taking a certain action in a certain state, and the Q value is a constant value;

[0084] S4.4. After each iteration, the immediate reward and new state are obtained after executing the action, the Q value is updated according to the Bellman equation, and the update of the constraint parameter optimization equation is introduced, that is, the action that maximizes the Q value is selected as the next action;

[0085] The optimized Bellman equation updates the Q value, where the learning rate determines the speed at which the algorithm updates the Q value. If the learning rate is set too high, the algorithm may jump too much when updating the Q value, resulting in an unstable learning process or even divergence. By introducing a limit parameter to optimize the update of the equation, limiting the size of the learning rate ensures that each update is not too aggressive, which helps the algorithm converge to the optimal strategy and helps the algorithm generalize better during the training process. Specifically:

[0086] Q(s,a)←Q(s,a)+Pα[R+γmax a′Q(s′, a′)-Q(s, a)];

[0087] Where Q(s, a) is the expected long-term return of taking action a to adjust the charge and discharge power of the battery cell in state space s, R is the immediate reward obtained immediately after taking the action, P is the restriction parameter, α is the learning rate, γ is the discount factor, s′ is the new state after taking the action, and a′ is the best action among all actions in the new state;

[0088] S4.5, using the Softmax strategy to balance exploration and exploitation, which allows the algorithm to find a balance between trying new actions and repeating actions with known high Q values, that is, selecting the action with the highest Q value with a certain probability;

[0089] S4.6. When the change in Q value becomes very small, that is, the algorithm no longer significantly updates the Q table, it can be considered that the learning process has converged. The strategy at this time is considered to be optimized and can effectively control the SOC of the battery cell and achieve balanced control.

[0090] Among them, the target optimization module considers minimizing SOC difference, maximizing battery life, and optimizing energy utilization efficiency through genetic algorithm to find the optimal balancing strategy, involving the following steps:

[0091] S6.1. Set up a multi-objective optimization framework, with the objectives including minimizing the SOC difference between battery cells, maximizing the battery pack life, and optimizing energy utilization efficiency;

[0092] S6.2, define multi-objective function using weighted sum method;

[0093] The weighted summation method is specifically:

[0094] F(x)=w1f1(x)+w2f2(x)+w3f3(x)

[0095] Among them, F(x) represents the weighted sum value, f1(x) represents minimizing the SOC difference between battery cells, f2(x) represents maximizing the battery pack life, f3(x) represents optimizing the energy utilization efficiency, and w1, w2, and w3 represent the target weights.

[0096] S6.3, select a genetic algorithm to solve the objective function;

[0097] The genetic algorithm is used to solve the function. By simulating the natural evolution process, the genetic algorithm can search in multiple areas of the solution space at the same time, which helps to jump out of the local optimal solution and find the global optimal solution. It can also process multiple solutions in the population at the same time and can run efficiently in a multi-core processor or distributed computing environment, greatly speeding up the search process. Specifically:

[0098] S6.31, generate W solutions as the initial population, each solution represents a SOC equilibrium strategy;

[0099] S6.32, calculate the fitness value of each solution by weighted sum method, that is, the performance of each strategy in satisfying the objective function;

[0100] S6.33, using tournament selection, select individuals from the current population to enter the breeding pool, and the probability of selection is proportional to the fitness of the individual;

[0101] S6.34, Generate new individuals through crossover and mutation operations. The crossover operation simulates the genetic process of organisms, while the mutation operation introduces new genetic mutations.

[0102] S6.35, replace old individuals with newly generated individuals to form a new population;

[0103] S6.36. If the quality of the solution reaches the predetermined standard A, analyze the solutions in the final population and select the optimal solution as the SOC equilibrium strategy.

[0104] S6.4. Evaluate the solution set generated by the algorithm and select the optimal solution that meets the multi-objective optimization criteria.

[0105] The real-time control system 2 uses the random forest algorithm to adjust and control the balancing control strategy in real time. The random forest algorithm can make accurate predictions on complex data distributions. In real-time control scenarios, this enables the random forest to accurately predict the SOC state and other key performance indicators of the battery cell, providing a basis for dynamically adjusting the control strategy. The diversified decision tree set reduces the risk of overfitting and increases the stability of the model. Even if the performance of individual battery cells fluctuates, the model can maintain a high prediction accuracy and reliability of the control strategy. The following steps are involved:

[0106] S9.1. Collecting operating data of the battery unit, the operating data including voltage, current, temperature and SOC parameters;

[0107] S9.2. Preprocess the data and use the collected data to train the random forest model;

[0108] S9.3. Select the parameters of the random forest, including the number of trees m, the maximum depth of the tree w, and the minimum number of samples s required for node splitting, and perform cross-validation to evaluate the performance of the model under different parameter combinations;

[0109] S9.4, collect real-time battery status data, input these data into the trained random forest model for prediction to predict the best control strategy;

[0110] Among them, in S9.4, the specific algorithm for inputting the real-time battery status data into the random forest model is:

[0111] S9.41. Use sample x to make predictions and get numerical predictions v b ;

[0112] S9.42. Calculate the average of the predicted values ​​to obtain the final prediction result:

[0113]

[0114] Among them, x represents the input real-time battery status data, v b It represents the SOC numerical result obtained after the input sample x is predicted, y represents the final prediction result of the input sample x, b represents an integer from 1 to B, and B represents the total number of predictions.

[0115] S9.5. Based on the prediction results of the model, the system will adjust the SOC balancing control strategy, which includes determining the charging and discharging sequence, charging and discharging rate, and whether passive control is required for each battery cell in the battery pack. Once a new control strategy is generated, the system will send instructions to the various components of the energy storage system to instruct them to operate according to the new strategy. The instructions include adjusting the output power of the charger or inverter, starting the battery balancing circuit, and adjusting the parameter settings of the battery management system (BMS).

[0116] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. The energy storage EMS system SOC balancing control system based on artificial intelligence is characterized by: include A monitoring system (4), wherein the monitoring system (4) collects voltage, current, temperature and SOC parameters of the battery cell; A balancing control system (1) is provided, wherein the balancing control system (1) balances and controls the SOC of each battery cell in the battery pack through a reinforcement learning algorithm RStudy, and then introduces the temperature data of the battery cell into a mean square error loss function optimized for battery temperature weights, and adjusts the contribution of the sample to the total loss function according to the temperature condition corresponding to the sample. If the prediction at a certain temperature is very important, then w i It will be set higher, so as to give a larger penalty to the prediction error of the sample when calculating the total loss, so as to achieve the accuracy of predicting the SOC of the battery cell under different temperature conditions; Among them, the optimized mean square error loss function is: Among them, w represents the weight, represents the weighted loss function, Y represents the actual SOC value, Represents the predicted SOC value, Y i represents the actual SOC value of the i-th sample, represents the predicted SOC value of the i-th sample, w i represents the temperature weight of the i-th sample; A real-time control system (2), wherein the real-time control system (2) uses a random forest algorithm to adjust the SOC balance control strategy in real time; A fault diagnosis and prevention system (3) is provided, wherein the fault diagnosis and prevention system (3) performs fault diagnosis through a random forest algorithm and takes preventive measures by adjusting a charging and discharging strategy.

2. The energy storage EMS system SOC balancing control system based on artificial intelligence according to claim 1 is characterized in that: The balanced control system (1) comprises a data prediction module, a dynamic adjustment module and a target optimization module; The data prediction module establishes the SOC prediction model of the battery cell through the long short-term memory network LSTM, analyzes the historical charge and discharge data of the battery cell and uses it as a reference to train the SOC prediction model. The trained SOC prediction model is then integrated into the real-time monitoring system to predict the SOC change trend of the battery cell by comparing the real-time monitoring data with the historical charge and discharge data. The dynamic adjustment module uses the reinforcement learning algorithm RStudy to dynamically adjust the actual SOC and predicted SOC of the battery cell, and realizes balanced control of SOC by adjusting the charge and discharge power of the battery cell, so that the actual SOC is as close as possible to the predicted SOC, thus achieving balanced control; The target optimization module finds the optimal SOC balance control strategy through genetic algorithm.

3. The SOC balancing control method of the energy storage EMS system based on artificial intelligence according to claim 2 is characterized in that: The data prediction module establishes a SOC prediction model for the battery cell through a long short-term memory network LSTM, involving the following steps: S3.

1. Collect the historical charge and discharge data of the battery cell and the corresponding SOC data to form a data set, which includes the voltage, current, and temperature parameters of the battery, and preprocess the data set to divide the data set into a training set, a validation set, and a test set. The training set is a data set used to train the LSTM model. The data set includes the voltage, current, temperature parameters of the battery, and the state of charge SOC of the battery cell. The LSTM model learns the mapping relationship between input and output through the training set. In this process, the LSTM model will try to minimize the mean square error loss function on the training set to learn how to make correct predictions. The validation set is used to adjust the hyperparameters of the LSTM model, which include the number of hidden layers in the neural network, the learning rate, etc., and to prevent overfitting. The test set is used to finally evaluate the generalization ability of the LSTM model. This is an independent evaluation of the model performance and does not participate in any training or hyperparameter tuning process. The results of the test set give the performance level that the LSTM model may achieve in practical applications. S3.

2. Use the deep learning framework TensorFlow to define the LSTM model architecture, and optimize the mean square error loss function by introducing battery temperature weights to achieve accurate prediction under different temperature conditions; S3.3, train the model using the training set data; S3.

4. Integrate the trained model into the real-time monitoring system and predict the SOC change trend based on the real-time monitoring data.

4. The SOC balancing control method of the energy storage EMS system based on artificial intelligence according to claim 2 is characterized in that: The dynamic adjustment module uses the reinforcement learning algorithm RStudy to achieve dynamic adjustment of the battery cell SOC, involving the following steps: S4.

1. Define a state space including the current SOC, predicted SOC, temperature, and charge and discharge state of the battery cell. The state set of each battery cell constitutes the state space of the overall system; define an action space for adjusting the charge and discharge power of the battery cell; S4.

2. Design a reward function that encourages energy balance between battery cells. When the SOC of a battery cell is close to the target value, a positive reward is given, otherwise a negative reward is given. S4.3, initialize the Q table, assign an initial Q value to each state-action combination in the state space and action space, that is, the sum of future rewards expected from taking a certain action in a certain state, and the Q value is a constant value; S4.

4. After each iteration, the immediate reward and new state are obtained after executing the action, the Q value is updated according to the Bellman equation, and the update of the constraint parameter optimization equation is introduced, that is, the action that maximizes the Q value is selected as the next action; S4.5, using the Softmax strategy to balance exploration and exploitation, which allows the algorithm to find a balance between trying new actions and repeating actions with known high Q values, that is, selecting the action with the highest Q value with a certain probability; S4.

6. When the change in Q value becomes very small, that is, the algorithm no longer significantly updates the Q table, it can be considered that the learning process has converged. The strategy at this time is considered to be optimized and can effectively control the SOC of the battery cell and achieve balanced control.

5. The SOC balancing control method of the energy storage EMS system based on artificial intelligence according to claim 4 is characterized in that: In S4.4, the optimized Bellman equation updates the Q value, where the learning rate determines the speed at which the algorithm updates the Q value. If the learning rate is set too high, the algorithm may jump too much when updating the Q value, resulting in an unstable learning process or even divergence. By introducing a limit parameter to optimize the update of the equation, limiting the size of the learning rate to ensure that each update is not too aggressive, it helps the algorithm converge to the optimal strategy and helps the algorithm generalize better during the training process. Specifically: Q(s,a)←Q(s,a)+Pα[R+γmax a′ Q(s′,a′)-Q(s,a)]; Where Q(s, a) is the expected long-term reward of taking action a to adjust the charge and discharge power of the battery cell in state space s, R is the immediate reward obtained immediately after taking the action, P is the restriction parameter, α is the learning rate, γ is the discount factor, s′ is the new state after taking the action, and a′ is the best action among all actions in the new state.

6. The SOC balancing control method of the energy storage EMS system based on artificial intelligence according to claim 2 is characterized in that: The target optimization module uses a genetic algorithm to consider minimizing SOC differences, maximizing battery life, and optimizing energy utilization efficiency to find the optimal balancing strategy, involving the following steps: S6.

1. Set up a multi-objective optimization framework, with the objectives including minimizing the SOC difference between battery cells, maximizing the battery pack life, and optimizing energy utilization efficiency; S6.2, define multi-objective function using weighted sum method; S6.3, select a genetic algorithm to solve the objective function; S6.

4. Evaluate the solution set generated by the algorithm and select the optimal solution that meets the multi-objective optimization criteria.

7. The SOC balancing control method of the energy storage EMS system based on artificial intelligence according to claim 6 is characterized by: In S6.2, the weighted summation method is specifically: F(x)=w1f1(x)+w2f2(x)+w3f3(x) Among them, F(x) represents the weighted sum value, f1(x) represents minimizing the SOC difference between battery cells, f2(x) represents maximizing the battery pack life, f3(x) represents optimizing the energy utilization efficiency, and w1, w2, and w3 represent the target weights.

8. The SOC balancing control method of the energy storage EMS system based on artificial intelligence according to claim 6 is characterized by: In S6.3, the genetic algorithm is used to solve the function: S6.31, generate W solutions as the initial population, each solution represents a SOC equilibrium strategy; S6.32, calculating the fitness value of each solution by weighted sum method; S6.33, using tournament selection, select individuals from the current population to enter the breeding pool, and the probability of selection is proportional to the fitness of the individual; S6.34, Generate new individuals through crossover and mutation operations. The crossover operation simulates the genetic process of organisms, while the mutation operation introduces new genetic mutations. S6.35, replace old individuals with newly generated individuals to form a new population; S6.

36. If the quality of the solution reaches the predetermined standard A, analyze the solutions in the final population and select the optimal solution as the SOC equilibrium strategy.

9. The SOC balancing control method of the energy storage EMS system based on artificial intelligence according to claim 1 is characterized by: The real-time control system (2) performs real-time adjustment and control on the equilibrium control strategy through a random forest algorithm, involving the following steps: S9.

1. Collecting operating data of the battery unit, the operating data including voltage, current, temperature and SOC parameters; S9.

2. Preprocess the data and use the collected data to train the random forest model; S9.

3. Select the parameters of the random forest, including the number of trees m, the maximum depth of the tree w, the minimum number of samples s required for node splitting, and perform cross-validation; S9.

4. Collect real-time battery status data and input the data into the trained random forest model for prediction; S9.

5. Adjust the balance control strategy in real time according to the prediction results of the model.

10. The SOC balancing control method of the energy storage EMS system based on artificial intelligence according to claim 9 is characterized in that: In S9.4, the specific algorithm for inputting the real-time battery status data into the random forest model is: S9.

41. Use sample x to make predictions and get numerical predictions v b ; S9.

42. Calculate the average of the predicted values ​​to obtain the final prediction result: Among them, x represents the input real-time battery status data, v b It represents the SOC numerical result obtained after the input sample x is predicted, y represents the final prediction result of the input sample x, b represents an integer from 1 to B, and B represents the total number of predictions.

Images