Lithium battery SOC estimation method based on ACO-GRU-LSTM

Abstract

The invention relates to a lithium battery SOC estimation method based on ACO-GRU-LSTM, and the method comprises the following steps: extracting current and voltage data key features through employing a singular value decomposition method, a moving mean value method, a moving variance method and a moving root-mean-square method, and carrying out the normalization processing; using an ant colony optimization algorithm (ACO) to optimize GRU-LSTM hyper-parameters (a hidden layer neuron number, a learning rate, a batch size and a Dropout rate); based on the optimized GRU-LSTM model, carrying out preliminary estimation on the SOC; and smoothing the SOC estimation result by using moving average filtering, and verifying the estimation precision through error analysis. Compared with the prior art, the SOC estimation method based on combination of ant colony optimization and GRU-LSTM has the advantages that the GRU-LSTM hyper-parameters can be optimized, the generalization ability of the model under different working conditions is improved, and high accuracy and stability are kept in SOC estimation.

Description

Technical Field

[0001] The present invention relates to the technical field of SOC estimation methods for ternary lithium power batteries, and in particular to a lithium battery SOC estimation method based on ACO-GRU-LSTM. Background Art

[0002] With the rapid development of the electric vehicle industry, lithium-ion batteries have become the main power source for electric vehicles due to their high energy density, long cycle life, and excellent charge and discharge characteristics. The SOC of a battery is an important indicator for measuring the remaining power and health of the battery, which directly affects the endurance and safety of electric vehicles. Accurate SOC estimation is essential to improving the performance of the battery management system, optimizing the battery life, and ensuring the safe operation of electric vehicles.

[0003] At present, SOC estimation methods mainly include: model-based methods (such as Kalman filtering, sliding mode observer), data-driven methods (such as neural networks, deep learning) and fusion methods. Among them, data-driven methods, especially recurrent neural network RNN ​​structures, such as long short-term memory networks LSTM and gated recurrent units GRU, have gradually become a research hotspot for SOC estimation due to their advantages in processing time series data. LSTM regulates information flow through forget gates, input gates and output gates, and is suitable for modeling long-term dependencies, while GRU uses update gates and reset gates to reduce computational complexity and speed up training. Therefore, combining the advantages of GRU and LSTM is expected to further improve the accuracy and stability of SOC estimation.

[0004] However, the performance of GRU-LSTM is highly dependent on the optimization of hyperparameters, such as the number of hidden layer neurons, learning rate, batch size, and dropout rate. Existing hyperparameter optimization methods, such as manual experience adjustment or grid search, have problems such as low optimization efficiency and easy to fall into local optimality.

[0005] To address this problem, it is necessary to develop an algorithm for automatically optimizing the key hyperparameters of GRU-LSTM, improve the generalization ability of the model, reduce the SOC estimation error, and enhance the adaptability of the battery management system under complex working conditions. Summary of the invention

[0006] The technical problem to be solved by the present invention is to automatically optimize GRU-LSTM hyperparameters, improve the generalization ability of the model, reduce the SOC estimation error, and provide a lithium battery SOC estimation method based on ACO-GRU-LSTM.

[0007] To achieve the above object, the present invention provides the following technical solutions: A lithium battery SOC estimation method based on ACO-GRU-LSTM, the method comprising the following steps: S1. Data preprocessing and feature extraction: According to the charging and discharging experimental data of the lithium battery, the moving mean, moving variance, moving root mean square and singular value decomposition methods are used to extract the moving mean, moving variance, moving root mean square and left singular vector key features of the current and voltage data, and normalization is performed to enhance the relevance of the features to the SOC estimation. The data includes the battery terminal voltage and current; S2. Ant Colony Optimization (ACO) algorithm is used to optimize the hyperparameters of GRU-LSTM. The ant colony optimization algorithm (ACO) is used to optimize the hyperparameters of GRU-LSTM. The ant colony explores the search space and gradually converges to the optimal hyperparameter combination to minimize the SOC estimation error of the model. S3. Train the GRU-LSTM optimized by the ant colony optimization algorithm (ACO), take the features extracted from the current and voltage of the lithium battery charge and discharge experiment as input, and SOC as output to construct a training data set; use the GRU-LSTM optimized by the ant colony optimization algorithm (ACO) to estimate SOC, and use the root mean square error (RMSE) as the loss function to adjust the network weights and improve the SOC prediction accuracy; S4, SOC estimation result optimization and output, apply sliding average filtering to the SOC estimation value output by GRU-LSTM for smoothing to reduce the error caused by short-term fluctuations; compare with the real SOC data to evaluate the estimation accuracy and stability, and finally output the optimized SOC estimation result; As a further technical solution of the present invention, in step S1, the moving mean, moving variance, moving root mean square and singular value decomposition method are used to extract the moving mean, moving variance, moving root mean square and left singular vector key features of the current and voltage data, and the step of normalizing them includes: S11. Use the moving mean, moving variance, moving root mean square and singular value decomposition methods to extract key features. The calculation formula is as follows: ; In the formula, is the sliding window size, is the time series voltage and current data, For the current moment, is the time series within the window, is the moving average value of voltage and current at the current moment; ; In the formula, is the sliding window size, is the time series voltage and current data, For the current moment, is the time series within the window, is the moving average value of voltage and current at the current moment, is the moving variance of voltage and current at the current moment; ; In the formula, is the sliding window size, is the time series voltage and current data, For the current moment, is the time series within the window, is the moving RMS value of voltage and current at the current moment; ; In the formula, is the left singular vector matrix, is the singular value matrix, is the right singular vector matrix, is the time series voltage and current data; S12. Use the zero mean normalization method to normalize the extracted key data features. The calculation formula is as follows: ; In the formula, To extract the key data features from the current and voltage data, is the sample mean of the data feature, is the sample standard deviation of the data feature, The data are normalized.

[0008] As a further technical solution of the present invention, in step 2), the step of optimizing the hyperparameters of GRU-LSTM using the ant colony optimization algorithm (ACO) includes: S21. Set the objective function: ; ; in, is the root mean square error, is the sample size, For the predicted indivual value, For the actual indivual value, are the hyperparameters of GRU-LSTM, including the number of hidden layer neurons (numHiddenUnits), learning rate (InitialLearnRate), batch size (MiniBatchSize), and Dropout rate (Dropout Rate); S22. Generate hyperparameter solution vector: Ants generate candidate solutions in the search space. The hyperparameter vector for ants is: ; In the formula, is the hyperparameter of GRU-LSTM, is the number of neurons in the hidden layer, is the learning rate, is the batch size, is the random dropout rate; Ant In the The selection probability during the round iteration is determined by the pheromone concentration: ; In the formula, is the specific value of a hyperparameter, is the current iteration number, For all possible hyperparameter combinations, is the index of all possible hyperparameters, For Ants Choosing Hyperparameters The probability of is the pheromone concentration, is the heuristic information, and Separately control the influence of pheromones and heuristic information; S23, Ant Colony Optimization (ACO) algorithm adopts global update rules, and pheromone is updated after each iteration: ; ; In the formula, is the current pheromone concentration, is the updated pheromone concentration, is the pheromone volatility coefficient, is the number of ants, For Ants Contribution to pheromone increment, is a constant, For Ants The error of the trained GRU-LSTM; S24, repeat steps S22-S22 for loop calculation, stop optimization after reaching the maximum number of iterations, and output the optimal hyperparameter combination.

[0009] As a further technical solution of the present invention, in step 3), the step of obtaining the SOC estimation value using the GRU-LSTM optimized by the ant colony optimization algorithm (ACO) includes: S31, GRU-LSTM models time series data through a gating mechanism, where GRU adjusts the influence of historical information through update gates and reset gates, and LSTM dynamically adjusts the information flow through forget gates, input gates, and output gates, so that the weights of the recurrent network change dynamically in time steps, thereby adapting to the integration scale of different time steps; In the process of information transmission, GRU controls the degree of dependence of the current state on the historical state through the update gate. The calculation formula is as follows: ; S32, GRU selectively forgets some historical information by resetting the gate to extract the key information at the current moment. The calculation formula is as follows: ; S33. Calculate the new candidate hidden state, combine the current input and past information, and update the GRU hidden state at the current moment. The formula is as follows: ; In the formula, Controls the current hidden state for the update gate Preserve past state , for Activation function, is the input at the current moment, is the weight matrix of the update gate, is the weight matrix of the candidate hidden state, is the bias of the update gate, To reset the gate bias, is the bias of the candidate hidden state, To reset the gate, is a candidate hidden state, is the hyperbolic tangent activation function; S34, LSTM selects historical information to be retained or forgotten through the forget gate. This process is composed of Function control, the calculation formula is as follows: ; In the formula, is the output of the forget gate, for Activation function, is the weight matrix of the forget gate, is the bias of the forget gate, is the hidden state at the previous moment, is the input at the current moment; S35, LSTM selects the information to be updated at the current moment through the input gate, and calculates the control variable of the input gate and the new candidate cell state, as shown in the following formula: ; ; In the formula, is the output of the input gate, is the weight matrix of the input gate, is the bias of the input gate, is the candidate cell state, is the weight matrix of candidate cells, Bias for candidate cells; S36. Calculate the cell state at the current moment: ; In the formula, is the cell state at the current moment, is the cell state at the previous moment; S37, LSTM determines the final output at the current moment through the output gate and updates the hidden state: ; ; In the formula, is the output gate, is the weight matrix of the output gate, is the bias of the output gate, is the hidden state at the current moment.

[0010] As a further technical solution of the present invention, in step S4, the step of applying a sliding average filter to smooth the SOC estimation value output by the GRU-LSTM includes: ; In the formula, is the smooth output of SOC at the current moment, For the current moment, is the offset of the time index, is the sliding window size, is the current moment and its previous The SOC estimation value is processed by sliding average filtering and smoothing to output the final estimated SOC value.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. The method of the present invention uses moving mean, moving variance, moving root mean square, and singular value decomposition to extract key features of current and voltage data, retain key information of experimental data, reduce the impact of noise, improve data stability, enhance the learning ability of neural networks, and improve SOC estimation accuracy.

[0012] 2. The method of the present invention combines the advantages of GRU and LSTM. GRU is used to efficiently process short-term dependencies, and LSTM is good at capturing long-term dependency information, making SOC estimation more accurate and stable.

[0013] 3. The method of the present invention uses the ant colony optimization algorithm (ACO) to adaptively optimize the GRU-LSTM hyperparameters, dynamically search for the optimal hyperparameter combination, and effectively reduce the estimation error of SOC.

[0014] 4. The method of the present invention uses a sliding average filter to improve the robustness of SOC prediction. After using GRU-LSTM to predict SOC, a sliding average filter is used to smooth the prediction results to reduce short-term fluctuations and improve the stability of SOC prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of current and voltage data feature extraction according to the present invention.

[0016] Figure 2 This is a flowchart of the combination of the ant colony optimization algorithm (ACO) and GRU-LSTM of the present invention.

[0017] Figure 3 This is a structural diagram of the monomer GRU unit of the present invention.

[0018] Figure 4 This is a structural diagram of a monomer LSTM unit of the present invention.

[0019] Figure 5 This is the structural diagram of the GRU-LSTM model of the present invention.

[0020] Figure 6 This is a comparison chart of the SOC estimation results of the ACO-GRU-LSTM and LSTM models in the example.

[0021] Figure 7 This is a comparison chart of the SOC estimation results for the mixed working condition in the example.

[0022] Figure 8 This is a comparison chart of the sliding filter processing results in the example. DETAILED DESCRIPTION

[0023] The technical solution of this patent is further described in detail below in conjunction with specific implementation methods.

[0024] The embodiment of the present invention provides a lithium battery SOC estimation method based on ACO-GRU-LSTM, comprising the following steps: S1. Data preprocessing and feature extraction: Based on the charging and discharging experimental data of the lithium battery, the data includes the battery terminal voltage and current. The moving mean, moving variance, moving root mean square and singular value decomposition methods are used to extract the key features of the current and voltage data, such as the moving mean, moving variance, moving root mean square and left singular vector, and normalization is performed to enhance the relevance of the features to the SOC estimation. The steps of extracting key features through moving mean, moving variance, moving root mean square and singular value decomposition and performing normalization processing include: S11. Use moving mean, moving variance, moving root mean square and singular value decomposition methods to extract key features. The process is as follows: Figure 1 shown.

[0025] ; In the formula, is the sliding window size, is the time series voltage and current data, For the current moment, is the time series within the window, is the moving average value of voltage and current at the current moment; ; In the formula, is the sliding window size, is the time series voltage and current data, For the current moment, is the time series within the window, is the moving average value of voltage and current at the current moment, is the moving variance of voltage and current at the current moment; ; In the formula, is the sliding window size, is the time series voltage and current data, For the current moment, is the time series within the window, is the moving RMS value of voltage and current at the current moment; ; In the formula, is the left singular vector matrix, is the singular value matrix, is the right singular vector matrix, is the time series voltage and current data; S12. Use the zero mean normalization method to normalize the extracted key data features. The calculation formula is as follows: ; In the formula, To extract the key data features from the current and voltage data, is the sample mean of the data feature, is the sample standard deviation of the data feature, The data are normalized.

[0026] S2. Ant Colony Optimization (ACO) optimizes GRU-LSTM hyperparameters. Ant Colony Optimization (ACO) is used to optimize GRU-LSTM hyperparameters. Ant colony explores the search space and gradually converges to the optimal hyperparameter combination to minimize the SOC estimation error of GRU-LSTM. The optimization process of Ant Colony Optimization (ACO) is as follows: Figure 2 As shown; The steps to optimize the hyperparameters of GRU-LSTM using the ant colony optimization (ACO) algorithm include: S21. Set the objective function: ; ; in, is the root mean square error, is the sample size, For the predicted indivual value, For the actual indivual value, are the hyperparameters of GRU-LSTM, including the number of hidden layer neurons (numHiddenUnits), learning rate (InitialLearnRate), batch size (MiniBatchSize), and Dropout rate (Dropout Rate); S22. Generate hyperparameter solution vector: Ants generate candidate solutions in the search space. The hyperparameter vector for ants is: ; In the formula, is the hyperparameter of GRU-LSTM, is the number of neurons in the hidden layer, is the learning rate, is the batch size, is the random dropout rate; Ant In the The selection probability during the round iteration is determined by the pheromone concentration: ; In the formula, is the specific value of a hyperparameter, is the current iteration number, For all possible hyperparameter combinations, is the index of all possible hyperparameters, For Ants Choosing Hyperparameters The probability of is the pheromone concentration, is the heuristic information, and Separately control the influence of pheromones and heuristic information; S23, Ant Colony Optimization (ACO) algorithm adopts global update rules, and pheromone is updated after each iteration: ; ; In the formula, is the current pheromone concentration, is the updated pheromone concentration, is the pheromone volatility coefficient, is the number of ants, For Ants Contribution to pheromone increment, is a constant, For Ants The error of the trained GRU-LSTM; S24, repeat steps S22-S23 to perform calculations in a loop, stop optimization after reaching the maximum number of iterations, and output the optimal hyperparameter combination.

[0027] S3. Train the GRU-LSTM network optimized by the ant colony optimization algorithm (ACO), use the current and voltage extracted from the lithium battery charge and discharge experiment as input, and SOC as output to construct a training data set; use the GRU-LSTM optimized by the ant colony optimization algorithm (ACO) to estimate SOC, and use the root mean square error (RMSE) as the loss function to adjust the network weights and improve the SOC prediction accuracy; The steps to obtain the SOC estimate using the GRU-LSTM optimized by the ant colony optimization algorithm (ACO) include: S31, GRU-LSTM models time series data through a gating mechanism, where GRU adjusts the influence of historical information through update gates and reset gates, and LSTM dynamically adjusts the information flow through forget gates, input gates, and output gates, so that the weights of the recurrent network change dynamically in time steps, thereby adapting to the integration scale of different time steps. The structure of a single GRU unit is as follows: Figure 3 As shown; In the process of information transmission, GRU controls the degree of dependence of the current state on the historical state through the update gate. The calculation formula is as follows: ; S32, GRU selectively forgets some historical information by resetting the gate to extract the key information at the current moment. The calculation formula is as follows: ; S33. Calculate the new candidate hidden state, combine the current input and past information, and update the GRU hidden state at the current moment. The formula is as follows: ; In the formula, Controls the current hidden state for the update gate Preserve past state , for Activation function, is the input at the current moment, is the weight matrix of the update gate, is the weight matrix of the candidate hidden state, is the bias of the update gate, To reset the gate bias, is the bias of the candidate hidden state, To reset the gate, is a candidate hidden state, is the hyperbolic tangent activation function; S34, LSTM selects historical information to be retained or forgotten through the forget gate. This process is controlled by the sigmoid function, and the calculation formula is as follows: ; In the formula, is the output of the forget gate, for Activation function, is the weight matrix of the forget gate, is the bias of the forget gate, is the hidden state at the previous moment, is the input at the current moment; S35, LSTM selects the information to be updated at the current moment through the input gate, and calculates the control variable of the input gate and the new candidate cell state, as shown in the following formula: ; ; In the formula, is the output of the input gate, is the weight matrix of the input gate, is the bias of the input gate, is the candidate cell state, is the weight matrix of candidate cells, Bias for candidate cells; S36. Calculate the cell state at the current moment: ; In the formula, is the cell state at the current moment, is the cell state at the previous moment; S37, LSTM determines the final output at the current moment through the output gate and updates the hidden state: ; ; In the formula, is the output gate, is the weight matrix of the output gate, is the bias of the output gate, is the hidden state at the current moment, and the structure of a single LSTM unit is as follows Figure 4 As shown, the GRU-LSTM model structure is as follows Figure 5 As shown; S4. Optimize and output the SOC estimation results. Apply sliding average filtering to smooth the SOC estimation value output by GRU-LSTM to reduce the error caused by short-term fluctuations. Compare it with the actual SOC data to evaluate the estimation accuracy and stability, and finally output the optimized SOC estimation result.

[0028] The steps of applying sliding average filtering to smooth the SOC estimate output by GRU-LSTM include: ; In the formula, is the smooth output of SOC at the current moment, For the current moment, is the offset of the time index, is the sliding window size, is the current moment and its previous The SOC estimation value is processed by sliding average filtering and smoothing to output the final estimated SOC value.

[0029] The lithium battery SOC estimation method based on ACO-GRU-LSTM of the present invention takes a single INR 18650-20R lithium battery as an example, uses DST and FUDS working condition experimental data for feature extraction at an ambient temperature of 25°C, uses DST working condition data for training, and uses the complete FUDS working condition for prediction. The ant colony optimization algorithm (ACO) optimizes GRU-LSTM to estimate SOC. Compared with LSTM, MAE and RMSE are reduced by 20.35% and 20.33% respectively. The estimation results are as follows: Figure 6 As shown. Under the mixed working condition of DST+FUDS, the ant colony optimization algorithm (ACO) optimizes GRU-LSTM to estimate SOC. Compared with LSTM, MAE and RMSE are reduced by 34.33% and 30.38% respectively. The estimation results are shown in Figure 7 After the SOC prediction result is processed by sliding filtering, the error caused by short-term fluctuations can be reduced, and the estimated curve is smoother, as shown in Figure 8 Therefore, the lithium battery SOC estimation method based on ACO-GRU-LSTM of the present invention improves the generalization ability of the model, reduces the SOC estimation error, and enhances the adaptability under complex working conditions.

[0030] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be considered exemplary and non-restrictive in all respects, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims be included in the present invention.

[0031] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A lithium battery SOC estimation method based on ACO-GRU-LSTM, characterized in that: The method comprises the following steps: S1. Data preprocessing and feature extraction. According to the charging and discharging experimental data of lithium batteries, the moving mean, moving variance, moving root mean square and singular value decomposition methods are used to extract the key features of the current and voltage data, such as the moving mean, moving variance, moving root mean square and left singular vector, and normalization is performed to enhance the relevance of the features to SOC estimation. The charging and discharging experimental data include the battery terminal voltage and current. S2. Ant colony optimization algorithm is used to optimize the hyperparameters of GRU-LSTM. The ant colony optimization algorithm is used to optimize the hyperparameters of GRU-LSTM. The ant colony explores the search space and gradually converges to the optimal hyperparameter combination to minimize the SOC estimation error of GRU-LSTM. S3. Train the GRU-LSTM optimized by the ant colony optimization algorithm, take the features extracted from the current and voltage of the lithium battery charge and discharge experiment as input, and SOC as output to construct a training data set; use the GRU-LSTM optimized by the ant colony optimization algorithm to estimate SOC, and use the root mean square error as the loss function to adjust the network weights and improve the SOC prediction accuracy; S4. Optimize and output the SOC estimation results. Apply sliding average filtering to smooth the SOC estimation value output by GRU-LSTM to reduce the error caused by short-term fluctuations. Compare it with the actual SOC data to evaluate the estimation accuracy and stability, and finally output the optimized SOC estimation result.

2. The lithium battery SOC estimation method based on ACO-GRU-LSTM according to claim 1, characterized in that: In step S1, the steps of extracting the key features of the moving mean, moving variance, moving root mean square, and left singular vector of the current and voltage data by using the moving mean, moving variance, moving root mean square, and singular value decomposition method, and performing normalization processing include: S11. Use the moving mean, moving variance, moving root mean square and singular value decomposition methods to extract key features. The calculation formula is as follows: ； In the formula, is the sliding window size, is the time series voltage and current data, For the current moment, is the time series within the window, is the moving average value of voltage and current at the current moment; ； In the formula, is the sliding window size, is the time series voltage and current data, For the current moment, is the time series within the window, is the moving average value of voltage and current at the current moment, is the moving variance of voltage and current at the current moment; ； In the formula, is the sliding window size, is the time series voltage and current data, For the current moment, is the time series within the window, is the moving RMS value of voltage and current at the current moment; ； In the formula, is the left singular vector matrix, is the singular value matrix, is the right singular vector matrix, is the time series voltage and current data; S12. Use the zero mean normalization method to normalize the moving mean, moving variance, moving root mean square, and left singular vector of the extracted current and voltage data. The calculation formula is as follows: ； In the formula, To extract the key data features from the current and voltage data, is the sample mean of the data feature, is the sample standard deviation of the data feature, The data are normalized.

3. The lithium battery SOC estimation method based on ACO-GRU-LSTM according to claim 1, characterized in that: In step S2, the steps of optimizing the hyperparameters of the GRU-LSTM using the ant colony optimization algorithm include: S21. Set the objective function: ； ； in, is the root mean square error, is the sample size, For the predicted indivual value, For the actual indivual value, are the hyperparameters of GRU-LSTM, including the number of hidden layer neurons, learning rate, batch size, and Dropout rate; S22. Generate hyperparameter solution vector: Ants generate candidate solutions in the search space. The hyperparameter vector for ants is: ； In the formula, is the hyperparameter of GRU-LSTM, is the number of neurons in the hidden layer, is the learning rate, is the batch size, is the random dropout rate; Ant In the The selection probability during the round iteration is determined by the pheromone concentration: ； In the formula, is the specific value of a hyperparameter, is the current iteration number, For all possible hyperparameter combinations, is the index of all possible hyperparameters, For Ants Choosing Hyperparameters The probability of is the pheromone concentration, is the heuristic information, and Separately control the influence of pheromones and heuristic information; S23, the ant colony optimization algorithm adopts a global update rule, and the pheromone is updated after each iteration: ； ； In the formula, is the current pheromone concentration, is the updated pheromone concentration, is the pheromone volatility coefficient, is the number of ants, For Ants Contribution to pheromone increment, is a constant, For Ants The error of the trained GRU-LSTM; S24, repeat steps S22-S23 to perform calculations in a loop, stop optimization after reaching the maximum number of iterations, and output the optimal hyperparameter combination.

4. The lithium battery SOC estimation method based on ACO-GRU-LSTM according to claim 1, characterized in that: In step S3, the steps of obtaining the SOC estimation value using the GRU-LSTM optimized by the ant colony optimization algorithm include: S31, GRU-LSTM models time series data through a gating mechanism, where GRU adjusts the influence of historical information through update gates and reset gates, and LSTM dynamically adjusts the information flow through forget gates, input gates, and output gates, so that the weights of the recurrent network change dynamically in time steps, thereby adapting to the integration scale of different time steps; In the process of information transmission, GRU controls the degree of dependence of the current state on the historical state through the update gate. The calculation formula is as follows: ； S32, GRU selectively forgets some historical information by resetting the gate to extract the key information at the current moment. The calculation formula is as follows: ； S33. Calculate the new candidate hidden state, combine the current input and past information, and update the GRU hidden state at the current moment. The formula is as follows: ； In the formula, Controls the current hidden state for the update gate Preserve past state , for Activation function, is the input at the current moment, is the weight matrix of the update gate, is the weight matrix of the candidate hidden state, is the bias of the update gate, To reset the gate bias, is the bias of the candidate hidden state, To reset the gate, is a candidate hidden state, is the hyperbolic tangent activation function; S34, LSTM selects historical information to be retained or forgotten through the forget gate. This process is composed of Function control, the calculation formula is as follows: ； In the formula, is the output of the forget gate, for Activation function, is the weight matrix of the forget gate, is the bias of the forget gate, is the hidden state at the previous moment, is the input at the current moment; S35, LSTM selects the information to be updated at the current moment through the input gate, and calculates the control variable of the input gate and the new candidate cell state, as shown in the following formula: ； ； In the formula, is the output of the input gate, is the weight matrix of the input gate, is the bias of the input gate, is the candidate cell state, is the weight matrix of candidate cells, Bias for candidate cells; S36. Calculate the cell state at the current moment: ； In the formula, is the cell state at the current moment, is the cell state at the previous moment; S37, LSTM determines the final output at the current moment through the output gate and updates the hidden state: ； ； In the formula, is the output gate, is the weight matrix of the output gate, is the bias of the output gate, is the hidden state at the current moment.

5. The lithium battery SOC estimation method based on ACO-GRU-LSTM according to claim 1, characterized in that: In step S4, the step of applying a sliding average filter to smooth the SOC estimation value output by the GRU-LSTM includes: ； In the formula, is the smooth output of SOC at the current moment, For the current moment, is the offset of the time index, is the sliding window size, is the current moment and its previous The SOC estimation value is processed by sliding average filtering and smoothing to output the final estimated SOC value.

Images