Lithium ion battery health state estimation method

Abstract

The invention discloses a lithium ion battery state-of-health estimation method, which comprises the following steps of: acquiring battery charge and discharge data, extracting a health feature (HF) related to battery aging, and performing correlation analysis and screening on the extracted HF and a battery state-of-health (SOH) by using a Pearson correlation coefficient method; carrying out dimensionality reduction on the HF data obtained by screening by adopting a principal component analysis method; and inputting the data after dimension reduction into a genetic algorithm (GA) to obtain an optimal data set. The method comprises the following steps: selecting an optimal data set, dividing the optimal data set into a test data set and a training data set according to different proportions, training a BiLSTM network by using data of the training set, optimizing parameters in the BiLSTM by using a Levy Flight improved grey wolf optimization algorithm (IGWO), and constructing a GA-IGWO-BiLSTM network estimation model. And inputting test set data into the constructed GA-IGWO-BiLSTM estimation model, verifying whether the model meets SOH estimation design requirements, if the model does not meet the requirements, re-constructing the SOH estimation model, and if the requirements are met, inputting battery data into the estimation model to obtain an SOH real-time estimation value.

Description

Technical Field

[0001] This invention belongs to the field of lithium-ion battery management technology and relates to a method for estimating the health status of lithium-ion batteries. Background Technology

[0002] Lithium-ion batteries have advantages such as high energy density, long cycle life, and low self-discharge rate, and are therefore widely used as power batteries for new energy vehicles. As the number of charge-discharge cycles increases, irreversible physicochemical reactions occur inside the battery, leading to gradual performance degradation and continuous internal aging. The state of health (SOH) of the battery is an important indicator that accurately reflects the degree of battery aging, and accurate estimation of battery SOH is crucial for ensuring the safe operation of the battery.

[0003] Researchers have conducted numerous studies on battery SOH estimation methods. The main methods for estimating battery SOH include experimental methods, model methods, and data-driven methods. Experimental and model methods have certain limitations in estimating battery SOH, while data-driven methods are currently the primary research approach for estimating battery SOH.

[0004] As the mainstream method for estimating battery state of harmonics (SOH), data-driven methods still have certain shortcomings. Regarding estimation performance, some data-driven methods fail to consider the comprehensiveness of the data they process, resulting in poor SOH estimation results. In terms of estimation efficiency, some data-driven methods pursue complexity and computational load at the expense of SOH estimation efficiency. Patent (CN113917337A) discloses a battery SOH estimation method based on charging data and an LSTM network. This method first obtains the original battery dataset, then performs data cleaning, normalization, and feature extraction to obtain the final input features, which are divided into training and test sets. Next, an LSTM network model is constructed, and the trained LSTM network model is optimized using an ant colony algorithm. Finally, the test set data is input into the optimized model to estimate the battery SOH. However, in this method, the LSTM network can only train the input data in one direction, resulting in poor SOH estimation performance. Patent (CN115808627A) discloses a method and apparatus for predicting the state of health (SOH) of lithium batteries. This method first acquires battery data, uses grey relational analysis to filter the data, and obtains correlated data. Next, it acquires the original battery data, decomposes the original data using ensemble empirical mode decomposition, and generates SOH component data and label data. Then, it adds an attention mechanism to the output of a BiLSTM network, followed by a dense layer for mapping, constructing a CNN-BiLSTM model based on the attention mechanism. Finally, it trains and tests the model using correlated data, SOH component data, and label data, generating an optimized model and estimating the SOH. However, this method involves multi-layer iterations of the BiLSTM network, increasing the computational cost and complexity of SOH estimation. Therefore, it is necessary to provide a lithium-ion battery health state estimation method to improve the effectiveness and efficiency of SOH estimation. Summary of the Invention

[0005] The problem addressed by this invention is to provide a method for estimating the state of health (SOH) of lithium-ion batteries. On one hand, it extracts and filters health features from battery charge and discharge data, performs correlation analysis and dimensionality reduction on the extracted HF data, and obtains the optimal dataset. A BiLSTM network that processes data bidirectionally is used instead of a unidirectional LSTM network, solving the problem of incomplete data processing affecting estimation performance. On the other hand, the improved Grey Wolf Optimization (IGWO) algorithm by Levy Flight is used to optimize the three parameters of learning rate, number of iterations, and hidden layer nodes in the BiLSTM network. This achieves a balance between the number of iteration layers, complexity, and estimation efficiency when optimizing the model, solving the problem of neglecting SOH estimation efficiency while only considering its own complexity and computational cost.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for estimating the health status of a lithium-ion battery, the specific steps of which are as follows:

[0007] S1: Acquire voltage, current, energy, and time data during battery charge-discharge cycles, and extract health characteristics highly correlated with battery aging by segment;

[0008] S2: Use the Pearson correlation coefficient method to calculate and screen the correlation between the extracted health features and the state of battery health (SOH);

[0009] S3: Principal component analysis (PCA) was used to reduce the dimensionality of the health characteristics selected by the Pearson correlation coefficient method;

[0010] S4: Input the dimensionality-reduced data into the genetic algorithm (GA) for further data optimization. Follow the steps of generating an initial population, calculating fitness, selection operation, crossover operation, expected value judgment, and mutation operation to obtain the optimal dataset.

[0011] S5: The optimal dataset obtained after PCA dimensionality reduction and GA optimization is divided into training set data and test set data according to different proportions each time, and is input into the LSTM network k times in total, where k is a natural number;

[0012] S6: Train a BiLSTM network using the training set data, and optimize the parameters in the BiLSTM using the Levy Flight improved Grey Wolf Optimization (IGWO) algorithm, including the learning rate, number of iterations, and hidden layer nodes. Construct a GA-IGWO-BiLSTM network estimation model. The optimization steps are as follows:

[0013] (1) Determine the parameters that need to be optimized in BiLSTM, namely the learning rate, the number of iterations, and the number of hidden layer nodes. Then determine the four initial parameters in GWO, namely the search space, the maximum number of iterations t, etc. max Dimension D and gray wolf population size N;

[0014] (2) Assign the position vector of each gray wolf in the population to the three parameters that BiLSTM needs to optimize, and calculate the fitness value of the position vector.

[0015] (3) Select the gray wolf positions corresponding to the three largest fitness values ​​in the population and assign them to the position vector. and

[0016] (4) Introducing Levy flight allows GWO to escape local optima, improving its global search capability and expanding its search range. Then, the position vector is updated. and To maximize its fitness, determine whether the maximum number of iterations has been reached. If the maximum number of iterations has been reached, end the update; otherwise, continue to repeat steps (2) and (3).

[0017] (5) After the iteration is completed, the optimal x will be determined. α The location vector is assigned to the learning rate, number of iterations, and hidden layer nodes of the BiLSTM network, and the model is trained and predicted according to these parameters.

[0018] S7: Train the forward LSTM and backward LSTM layers separately using the training dataset, and construct the GA-IGWO-BiLSTM network estimation model. The BiLSTM network consists of two single-layer LSTM networks with opposite directions. One LSTM network processes the input data from the forward direction, and the other LSTM network processes the input data from the backward direction. After processing, the outputs of the two LSTM networks are concatenated and then the total output is calculated.

[0019] S8: Input the test set data into the GA-IGWO-BiLSTM estimation model to verify whether the model accuracy meets the expected design requirements. If it does not meet the requirements, rebuild the estimation model. If it does meet the requirements, input the battery voltage and current online data into the network model to perform real-time SOH estimation, thereby obtaining the battery SOH estimation values ​​under different proportions of training set data and test set data.

[0020] Preferably, the health characteristics mentioned in step S1 include dividing the constant current stage charging voltage range into different voltage intervals by taking a time interval Δt = t1, and extracting the constant current charging time T within the constant current charging interval. I Constant current charging energy E I Rated voltage V and voltage V after time interval Δt t The difference ΔV is used to divide the constant voltage charging current range into different current intervals by taking the time interval Δt = t2, and the constant voltage charging time T within the constant voltage charging interval is extracted. V Constant voltage charging energy E V Rated current I and current I after time interval Δt t The difference ΔI; T when Δt=t1 is within the charging voltage range of the constant current stage (1) I E I , ΔV and (2) T when Δt=t2 in the constant voltage stage charging current range V E V ΔI is a health characteristic.

[0021] Preferably, the Pearson correlation coefficient method used in step S2 is calculated using the following formula:

[0022]

[0023] In the formula, r represents the Pearson correlation coefficient, and x i and y i These represent health characteristic data and actual SOH values, respectively. and is the average of the two values, and n is the number of battery charge-discharge cycles.

[0024] Preferably, in step S3, principal component analysis (PCA) is used to reduce the dimensionality of health features. The specific steps are as follows: given an m×n sample feature matrix X, the sample feature matrix is ​​standardized to obtain X′, and then the covariance matrix C is calculated. According to the formula Cu i =λ i u i Calculate the eigenvector u of X′ i and eigenvalues ​​λ i Let i = 1, 2, ..., k. Then we obtain the dimension-reduced matrix H = X′ × U, where U = [u1, u2, ..., u...]. k ], using the formula Calculate the variance contribution rate of each principal component, and take the principal component corresponding to the feature value with a cumulative contribution rate of more than 80% as the new health feature.

[0025] Preferably, in step S4, a genetic algorithm (GA) is used for further optimization, following a series of steps including generating an initial population, calculating fitness, selection operation, crossover operation, expected value judgment, and mutation operation, to obtain the optimal dataset.

[0026] Preferably, in step S6, a Gray Wolf Optimization Algorithm (GWO) is designed to simulate the social hierarchy and predation behavior of gray wolf packs in nature. Wolf packs are social animals with a strict hierarchy. The α wolf is the highest leader, followed by β wolves, δ wolves, and ω wolves. The position of the α wolf is considered the optimal solution, the positions of the β and δ wolves are considered the good and second-best solutions, respectively, and the position of the ω wolf is considered a candidate solution. The α, β, and δ wolves guide the optimization, while the ω wolf follows the first three types of wolves. The gray wolf pack's hunting process includes three stages: surrounding the prey, hunting, and attacking or searching for prey, ultimately obtaining a local optimum, wherein:

[0027] (1) Encircling the prey: During the encirclement, each wolf is a potential solution, and α, β, and δ represent the three best solutions. Other individuals continuously update their positions based on these three optimal solutions to complete the optimization process. The mathematical formula for the gray wolf encirclement process is as follows: In the formula, t is the current iteration number. The iteration process decreases linearly from 2 to 0, t max Represents the maximum number of iterations. Let be the prey position vector in the t-th iteration. Let be the gray wolf's position vector in the t-th iteration. It is a coefficient vector. It is a distance vector. and It is a random vector between [0,1]. It is a random vector between [0, 2].

[0028] (2) Hunting: Hunting is guided by the α wolf, with β and δ wolves also participating. The α, β, and δ wolves are usually in the position closest to the prey. The other wolves will update their positions based on the positions of the three best wolves. The position update process is as follows:

[0029]

[0030] In the formula, This represents the position vectors of α wolf, β wolf, and δ wolf in the current iteration. Let represent the distances between the current gray wolf and the three optimal wolves, respectively. This represents the optimal solution vector at the (t+1)th iteration.

[0031] (3) Attacking or searching for prey: Attacking the prey determines its location, thus obtaining the optimal solution, according to the formula It can be seen that random vectors Selection and Regarding the value, it takes values ​​in the interval [-a, a], when At that time, the gray wolves conducted a global search, dispersing to various areas to continue searching for prey. At that time, the gray wolf searches for prey in a nearby area and attacks the prey, finding a local optimum.

[0032] Preferably, in step S6, Levy Flight is introduced to improve the GWO algorithm. Levy Flight simulates the random walking of organisms in nature, optimizing the hunting methods and ranges of α, β, and δ wolves in the GWO algorithm, allowing them to escape local optima. The random step size of the Levy distribution is determined by the formula... Calculations are performed where ν and μ follow a standard normal distribution. σ ν =1, Γ is the gamma function, representing the integer factorial of N in the range N-1 to 0, and θ typically takes the value 1.5; the optimal solution vector obtained after the update is in, η is the weight that controls the step size. Let represent the Hadamard product, and Levy be the random step size.

[0033] Preferably, in step S7, the internal structure of the LSTM consists of three parts: a forget gate, an input gate, and an output gate, thereby achieving information protection and control. BiLSTM is composed of a forward LSTM network and a backward LSTM network, which overcomes the limitation of traditional LSTM training which only propagates from front to back. After processing, the outputs of the two LSTM networks are concatenated before the final output. The concatenation method is as follows: the forward hidden output state sequence obtained through the forward LSTM network is: The reverse hidden output state sequence obtained by the reverse LSTM network: Merging the two hidden states yields... And output the final result, where: λ t ρ is the hidden layer weight of the forward output at time t. t b is the hidden layer weights output in reverse at time t. t H is the bias parameter at time t. t It is the output value of BiLSTM at time t.

[0034] Preferably, in step S7, mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) are selected as evaluation indicators to determine whether the design requirements are met. The specific formula is as follows: In the formula, N represents the estimated number of experiments, and y e This is the actual value of SOH. Let e ​​be the estimated value of SOH, where e = 1, 2, ..., N.

[0035] Compared with existing technologies, this invention has the following technical advantages: First, it uses a BiLSTM network that processes data bidirectionally to replace the LSTM network that processes data unidirectionally, reducing the accuracy problem caused by incomplete data processing and enabling more comprehensive processing of the optimal input dataset, thus improving the estimation effect of SOH. Second, it uses the improved Grey Wolf Optimization (IGWO) algorithm by Levy Flight to optimize the parameters in the BiLSTM network, reducing the number of iteration layers and complexity of the algorithm when optimizing the model, thereby improving the estimation efficiency of SOH. Attached Figure Description

[0036] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0037] In the attached diagram:

[0038] Figure 1 This is a flowchart of the lithium-ion battery health status estimation method of the present invention;

[0039] Figure 2This is a schematic diagram of the LSTM network model structure in this invention.

[0040] Figure 3 This is a schematic diagram of the BiLSTM network model structure in this invention. Detailed Implementation

[0041] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0042] Example: Figure 1 As shown, a method for estimating the health status of a lithium-ion battery includes the following steps:

[0043] S1: Acquire voltage, current, energy, and time data during the battery charge-discharge cycle, and extract health characteristics highly correlated with battery aging in different intervals. The battery charge-discharge cycle process is as follows: First, charge with a constant current (CC) of 1.5A until the voltage rises to 4.2V, then charge with a constant voltage (CV) until the current drops to 20mA. During the discharge process, maintain a constant current of 2A. Discharge ends when the battery voltage drops from 4.2V to the cutoff voltage. The extracted health characteristics are the constant current charging time T within the CC stage charging voltage range of 3.9V to 4.2V, with a time interval Δt = 30s. I Constant current charging energy E I Rated voltage V and voltage V after time interval Δt t The difference ΔV, the constant voltage charging time T within the charging current range of 1.5A to 0.5A during the CV stage, is taken as Δt = 15s. V Constant voltage charging energy E V Rated current I and current I after time interval Δt t The difference ΔI; T when Δt = 30s during the constant current stage charging voltage range (1) I E I , ΔV and (2) T when Δt=15s in the constant voltage stage charging current range V E V ΔI is a health characteristic.

[0044] S2: Calculate the correlation between the extracted health features and the state of battery health (SOH) using the Pearson correlation coefficient method and then filter the data, selecting health features with a correlation coefficient above 0.85. The Pearson correlation coefficient method used is calculated using the following formula:

[0045]

[0046] In the formula, r represents the Pearson correlation coefficient, and x i and yi These represent health characteristic data and actual SOH values, respectively. and is the average of the two values, and n is the number of battery charge-discharge cycles.

[0047] S3: Principal Component Analysis (PCA) is used to reduce the dimensionality of the health features selected by the Pearson correlation coefficient method. The specific steps are as follows: Given an m×n sample feature matrix X, the sample feature matrix is ​​standardized to obtain X′, and then the covariance matrix C is calculated. According to the formula Cu i =λ i u i Calculate the eigenvector u of X′ i and eigenvalues ​​λ i Let i = 1, 2, ..., k. Then we obtain the dimension-reduced matrix H = X′ × U, where U = [u1, u2, ..., u...]. k ], using the formula Calculate the variance contribution rate of each principal component, and take the principal component corresponding to the feature value with a cumulative contribution rate of more than 80% as the new health feature.

[0048] S4: Input the dimensionality-reduced data into the genetic algorithm (GA) for further data optimization. Following the steps of generating an initial population, calculating fitness, selection, crossover, predictive value judgment, and mutation, the optimal dataset is obtained.

[0049] S5: The optimal dataset obtained after PCA dimensionality reduction and GA optimization is divided into training set data and test set data according to different proportions each time, and is input into the LSTM network k times in total, where k is a natural number.

[0050] S6: Train a BiLSTM network using the training set data, and optimize the parameters in the BiLSTM using the Levy Flight improved Grey Wolf Optimization (IGWO) algorithm, including the learning rate, number of iterations, and hidden layer nodes. Construct a GA-IGWO-BiLSTM network estimation model. The optimization steps are as follows:

[0051] (1) Determine the parameters that need to be optimized in BiLSTM, namely the learning rate, the number of iterations, and the number of hidden layer nodes. Then determine the four initial parameters in GWO, namely the search space, the maximum number of iterations t, etc. max Given the dimension D and the gray wolf population size N, and setting the search space to the entire optimal dataset, the maximum number of iterations t is determined. max =500, dimension D=3, gray wolf population N=30;

[0052] (2) Assign the position vector of each gray wolf in the population to the three parameters that BiLSTM needs to optimize, and calculate the fitness value of the position vector.

[0053] (3) Select the gray wolf positions corresponding to the three largest fitness values ​​in the population and assign them to the position vector. and

[0054] (4) Introducing Levy flight allows GWO to escape local optima, improving its global search capability and expanding its search range. Then, the simulation step size is set to 1000, and the position vector is updated. and To maximize its fitness, determine if the maximum number of iterations has been reached. If the maximum number of iterations t has been reached... max =500, end the update; otherwise, continue to repeat steps (2) and (3).

[0055] (5) After the iteration is completed, the optimal x will be determined. α The location vector is assigned to the learning rate, number of iterations, and hidden layer nodes of the BiLSTM network, and the model is trained and predicted according to these parameters.

[0056] In step (1), a Grey Wolf Optimization Algorithm (GWO) is designed to simulate the social hierarchy and predation behavior of grey wolf packs in nature. The grey wolf hierarchy is as follows: wolves are social animals with a strict internal hierarchy. The α wolf is the highest leader, followed by the β wolf, δ wolf, and ω wolf. The position of the α wolf is considered the optimal solution, the positions of the β and δ wolves are considered the good and second-best solutions, respectively, and the position of the ω wolf is considered a candidate solution. The α, β, and δ wolves guide the optimization, while the ω wolf follows the first three types of wolves.

[0057] 1. Encircling Prey: During the encirclement process, each wolf represents a potential solution, and α, β, and δ represent the three best solutions. Other individuals continuously update their positions based on these three optimal solutions to achieve the best possible outcome. The mathematical formula for the gray wolf encirclement process is as follows: In the formula, t is the current iteration number. The iteration process decreases linearly from 2 to 0, t max Represents the maximum number of iterations. Let be the prey position vector in the t-th iteration. Let be the gray wolf's position vector in the t-th iteration. It is a coefficient vector. It is a distance vector. and It is a random vector between [0,1]. It is a random vector between [0, 2].

[0058] 2. Hunting: Hunting is guided by the alpha wolf, with beta and delta wolves also participating. The alpha, beta, and delta wolves are usually positioned closest to the prey. The other wolves update their positions based on the positions of the three best wolves. The position update formula is as follows: In the formula, This represents the position vectors of α wolf, β wolf, and δ wolf in the current iteration. Let represent the distances between the current gray wolf and the three optimal wolves, respectively. This represents the optimal solution vector at the (t+1)th iteration.

[0059] 3. Attacking or searching for prey: Attacking the prey determines its location, thus obtaining the optimal solution, according to the formula... It can be seen that random vectors Selection and Regarding the value, it takes values ​​in the interval [-a, a], when When the wolf pack moves away from the optimal solution, it performs a global search. When the wolf pack approaches the optimal solution, it performs a local search and finds a local optimum.

[0060] In step (4), Levy flight is introduced to improve the GWO algorithm. Levy flight simulates the random walking of organisms in nature, optimizing the hunting methods and ranges of α, β, and δ wolves in the GWO algorithm, so that they can escape local optima; the random step size of the Levy distribution is determined by the formula. Calculations are performed where ν and μ follow a standard normal distribution. σ ν =1, Γ is the gamma function, representing the integer factorial of N in the range N-1 to 0, and θ typically takes the value 1.5; the optimal solution vector obtained after the update is in, η is the weight that controls the step size. This represents the Hadamard product, where Levy is the random step size;

[0061] S7: Train the forward LSTM and backward LSTM layers separately using the training dataset, and construct the GA-IGWO-BiLSTM network estimation model. Each BiLSTM layer is composed of two single-layer LSTM layers operating in opposite directions. One LSTM layer processes the input data from the forward direction, and the other from the backward direction. After processing, the outputs of the two LSTM layers are concatenated for the final output.

[0062] The internal structure of an LSTM consists of three parts: a forget gate, an input gate, and an output gate, which enables information protection and control. A BiLSTM is composed of a forward LSTM network and a backward LSTM network, overcoming the limitation of traditional LSTMs which only propagate training from front to back. After processing, the outputs of the two LSTM layers are concatenated before the final output. The concatenation method is as follows: the forward hidden output state sequence obtained through the forward LSTM network is: The reverse hidden output state sequence obtained by the reverse LSTM network: Merging the two hidden states yields... And output the final result, where: λ t ρ is the hidden layer weight of the forward output at time t. t b is the hidden layer weights output in reverse at time t. t H is the bias parameter at time t. t It is the output value of BiLSTM at time t.

[0063] S8: Input the test set data into the GA-IGWO-BiLSTM estimation model to verify whether the model accuracy meets the expected design requirements. The mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) are selected as evaluation indicators to determine whether the design requirements are met. The specific formula is as follows: In the formula, N represents the estimated number of experiments, and y e This is the actual value of SOH. The SOH estimate is given by e = 1, 2, ..., N. If the condition is not met, the estimation model is reconstructed. If the condition is met, the online voltage and current data of the battery are input into the network model to perform real-time SOH estimation, thereby obtaining the battery SOH estimate under different proportions of training set data and test set data.

[0064] Finally, it should be noted that the above descriptions are merely examples of the present invention and are not intended to limit the invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for estimating the state of health of a lithium-ion battery, characterized in that, The method includes the following steps: S1: Acquire voltage, current, energy, and time data during battery charge-discharge cycles, and extract health characteristics highly correlated with battery aging by segment; S2: Use the Pearson correlation coefficient method to calculate and screen the correlation between the extracted health features and the state of battery health (SOH); S3: Principal component analysis (PCA) was used to reduce the dimensionality of the health characteristics selected by the Pearson correlation coefficient method; S4: Input the dimensionality-reduced data into the genetic algorithm (GA) for further data optimization. Follow the steps of generating an initial population, calculating fitness, selection operation, crossover operation, expected value judgment, and mutation operation to obtain the optimal dataset. S5: The optimal dataset obtained after PCA dimensionality reduction and GA optimization is divided into training set data and test set data according to different proportions each time, and is input into the LSTM network k times in total, where k is a natural number; S6: Train a BiLSTM network using the training set data, and optimize the parameters in the BiLSTM using the Levy Flight improved Grey Wolf Optimization (IGWO) algorithm, including the learning rate, number of iterations, and hidden layer nodes. Construct a GA-IGWO-BiLSTM network estimation model. The optimization steps are as follows: (1) Determine the parameters that need to be optimized in BiLSTM, namely the learning rate, the number of iterations, and the number of hidden layer nodes. Then determine the four initial parameters in GWO, namely the search space, the maximum number of iterations t, etc. max Dimension D and gray wolf population size N; (2) Assign the position vector of each gray wolf in the population to the three parameters that BiLSTM needs to optimize, and calculate the fitness value of the position vector. (3) Select the gray wolf positions corresponding to the three largest fitness values ​​in the population and assign them to the position vector. and (4) Introducing Levy flight allows GWO to escape local optima, improving its global search capability and expanding its search range. Then, the position vector is updated. and To maximize its fitness, determine whether the maximum number of iterations has been reached. If the maximum number of iterations has been reached, end the update; otherwise, continue to repeat steps (2) and (3). (5) After the iteration is completed, the optimal x will be determined. α The position vector is assigned to the learning rate, number of iterations, and hidden layer nodes of the BiLSTM network, and the model is trained and predicted according to these parameters. S7: Train the forward LSTM and backward LSTM layers separately using the training dataset, and construct the GA-IGWO-BiLSTM network estimation model. The BiLSTM network consists of two single-layer LSTM networks with opposite directions. One LSTM network processes the input data from the forward direction, and the other LSTM network processes the input data from the backward direction. After processing, the outputs of the two LSTM networks are concatenated and then the total output is calculated. S8: Input the test set data into the GA-IGWO-BiLSTM estimation model to verify whether the model accuracy meets the expected design requirements. If it does not meet the requirements, rebuild the estimation model. If it does meet the requirements, input the battery voltage and current online data into the network model to perform real-time SOH estimation, thereby obtaining the battery SOH estimation values ​​under different proportions of training set data and test set data.

2. The method for estimating the health status of a lithium-ion battery according to claim 1, characterized in that: The health characteristics mentioned in step S1 include dividing the constant current stage charging voltage range into different voltage intervals by taking the time interval Δt = t1, and extracting the constant current charging time T within the constant current charging interval. I Constant current charging energy E I Rated voltage V and voltage V after time interval Δt t The difference ΔV is used to divide the constant voltage charging current range into different current intervals by taking the time interval Δt = t2, and the constant voltage charging time T within the constant voltage charging interval is extracted. V Constant voltage charging energy E V Rated current I and current I after time interval Δt t Difference ΔI; (1) T when Δt=t1 is within the constant current stage charging voltage range. I E I , ΔV and (2) T when Δt=t2 in the constant voltage stage charging current range V E V ΔI is a health characteristic.

3. The method for estimating the state of health of a lithium-ion battery according to claim 1, characterized in that: The Pearson correlation coefficient method used in step S2 is calculated using the following formula: In the formula, r represents the Pearson correlation coefficient, and x i and y i These represent health characteristic data and actual SOH values, respectively. and is the average of the two values, and n is the number of battery charge-discharge cycles.

4. The method for estimating the health status of a lithium-ion battery according to claim 1, characterized in that: In step S3, principal component analysis (PCA) is used to reduce the dimensionality of health features. The specific steps are as follows: given an m×n sample feature matrix X, the sample feature matrix is ​​standardized to obtain X′, and then the covariance matrix C is calculated. According to the formula Cu i =λ i u i Calculate the eigenvector u of X′ i and eigenvalues ​​λ i Let i = 1, 2, ..., k. Then we obtain the dimension-reduced matrix H = X′ × U, where U = [u1, u2, ..., u...]. k ], using the formula Calculate the variance contribution rate of each principal component, and take the principal component corresponding to the feature value with a cumulative contribution rate of more than 80% as the new health feature.

5. The method for estimating the health status of a lithium-ion battery according to claim 1, characterized in that: In step S4, a genetic algorithm (GA) is used for further optimization. Following a series of steps, including generating an initial population, calculating fitness, selection, crossover, determining expected values, and mutation, the optimal dataset is obtained.

6. The method for estimating the health status of a lithium-ion battery according to claim 1, characterized in that: In step S6, the Grey Wolf Optimization Algorithm (GWO) is designed to simulate the social hierarchy and predation behavior of grey wolf packs in nature. Wolves are social animals with a strict hierarchy. The α wolf is the highest leader, followed by the β wolf, δ wolf, and ω wolf. The position of the α wolf is considered the optimal solution, the positions of the β and δ wolves are considered the good and second-best solutions, and the position of the ω wolf is considered the candidate solution. The α, β, and δ wolves guide the optimization, while the ω wolf follows the first three types of wolves. The grey wolf pack's hunting process includes three stages: surrounding the prey, hunting, and attacking or searching for the prey, ultimately obtaining a local optimum.

7. The method for estimating the health status of a lithium-ion battery according to claim 1, characterized in that: In step S6, Levy Flight is introduced to improve the GWO algorithm. Levy Flight simulates the random walking of organisms in nature and optimizes the hunting methods and ranges of α, β and δ wolves in the GWO algorithm, so that it can escape local optima.

8. The method for estimating the state of health of a lithium-ion battery according to claim 1, characterized in that: In step S7, the internal structure of the LSTM consists of three parts: a forget gate, an input gate, and an output gate, thereby achieving information protection and control. BiLSTM is composed of a forward LSTM network and a backward LSTM network, which overcomes the limitation of traditional LSTM training which only propagates from front to back. After processing, the outputs of the two LSTM layers are concatenated before the final output. The concatenation method is as follows: the forward hidden output state sequence obtained through the forward LSTM network is: The reverse hidden output state sequence obtained by the reverse LSTM network: Merging the two hidden states yields... And output the final result, where: λ t ρ is the hidden layer weight of the forward output at time t. t b is the hidden layer weights output in reverse at time t. t H is the bias parameter at time t. t It is the output value of BiLSTM at time t.

9. The method for estimating the health status of a lithium-ion battery according to claim 1, characterized in that: In step S7, mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) are selected as evaluation indicators to determine whether the design requirements are met. The specific formula is as follows: In the formula, N represents the estimated number of experiments, and y e This is the actual value of SOH. Let e ​​be the estimated value of SOH, where e = 1, 2, ..., N.

Images