Novel energy storage battery module health state multi-stage evaluation method

Abstract

The invention discloses a novel energy storage battery module health state multi-stage evaluation method, and belongs to the field of energy storage battery management, and the method comprises the following steps: firstly, measuring the capacity of a battery cell through a variable ampere integral method, and extracting a health factor through capacity increment analysis to represent battery attenuation; thirdly, training a battery monomer capacity estimation model based on the health factors and the capacity label of battery aging; and finally, when the health state of the battery pack is evaluated, the health state of the battery pack is evaluated by combining the battery cell capacity estimation value and the inconsistency characteristic in the battery pack, the overall prediction accuracy is improved, the aging trend of a single battery cell is captured, the potential problem of the battery pack is identified through the consistency characteristic of the battery pack, and the battery pack health state is evaluated. The method has high precision and practical value, and is especially suitable for health monitoring and life management of a large-capacity energy storage system.

Description

Technical Field

[0001] The present invention belongs to the field of energy storage battery management, and particularly relates to a multi-stage evaluation method for the health state of a new type of energy storage battery module. Background Art

[0002] Lithium-ion battery energy storage systems have been widely used in fields such as power grid peak shaving and frequency modulation, renewable energy grid connection and consumption, and distributed generation due to their advantages such as high energy density, low self-discharge rate, and long cycle life.

[0003] Currently, the battery SOH evaluation methods mainly include the following categories: One is the direct evaluation method based on capacity testing, which requires a complete charge and discharge cycle for the battery pack, but this method is time-consuming and will affect the normal operation of the system; the second is the indirect evaluation method based on internal resistance measurement. Although this method is convenient for testing, its accuracy is greatly affected by factors such as environmental temperature and measurement conditions; the third is the model estimation method based on operation data, such as using machine learning and deep learning algorithms to establish an SOH prediction model. However, the existing model estimation methods require a large amount of historical data support and are difficult to accurately reflect the real-time attenuation state of the battery pack. At the same time, for large-capacity energy storage systems, the existing methods generally lack consideration of the consistent attenuation of the battery pack. Summary of the Invention

[0004] (1) Technical Problems to be Solved

[0005] In order to overcome the situation that the existing battery SOH evaluation methods are time-consuming and affect the normal operation of the system, or the accuracy is greatly affected by factors such as environmental temperature and measurement conditions, or the existing model estimation methods require a large amount of historical data support and are difficult to accurately reflect the real-time attenuation state of the battery pack, and for large-capacity energy storage systems, the existing methods generally lack consideration of the consistent attenuation of the battery pack.

[0006] (2) Technical Solutions

[0007] The present invention proposes a multi-stage evaluation method for the health state of a new type of energy storage battery module. The method includes the following steps: First, use the variable ampere integration method to measure the cell capacity, and extract the health factor through capacity increment analysis to characterize the battery attenuation; then, based on these health factors and the capacity label of battery aging, train the battery single-cell capacity estimation model; finally, when evaluating the health state of the battery pack, combine the cell capacity estimation value with the inconsistency characteristics in the battery pack to evaluate the health state of the battery pack, improving the overall prediction accuracy. This method not only captures the aging trend of a single cell but also identifies potential problems in the battery pack through the battery pack consistency characteristics, has high accuracy and practical value, and is particularly suitable for the health monitoring and life management of large-capacity energy storage systems;

[0008] Specifically, it is implemented through the following technical solutions: including the following steps:

[0009] S1: Extract the label capacity and health factor based on the charging data of the energy storage battery, specifically including:

[0010] S11: Through the cycle aging experiment of the battery monomer, obtain the external characteristic data of the battery during the charge and discharge process, including key information such as charge / discharge time, voltage, current, state of charge (SOC), temperature, etc. When obtaining the capacity label, first set a fixed SOC interval of 30%-80% to avoid the influence of extreme sections on the capacity evaluation and ensure that the same battery operating state is covered in each calculation. Based on the variable ampere integration method, calculate the capacity label of the battery. The expression of the variable ampere integration method is:

[0011]

[0012] Among them, C a is the calculated battery capacity, Δt is the sampling interval, I is the charging current, t 1 and t 2 are the start and end times of discharge respectively, SOC t1 is the start SOC data of discharge, and SOC t2 is the end SOC data of discharge;

[0013] S12: Obtain the charging IC curves of the battery and extract a large number of health factors based on these curves. Different characteristics of the IC curves, such as slope, curve shape, spacing, etc., can reflect the internal electrochemical behavior and health status of the battery. In order to denoise and smooth the data in the IC curves, Gaussian filtering is used to reduce the interference of noise on feature extraction. The calculation formula of the IC curve is as follows:

[0014]

[0015] Among them, Q is the charging amount, V is the voltage, I is the current, and t 1 is the start time of the segment charging, and t 2 is the cut-off time of the segment;

[0016] The calculation formula of Gaussian filtering is as follows:

[0017]

[0018] Among them, G(x) is the Gaussian kernel function, x is the relative position in the filtering window, σ is the standard deviation of the Gaussian function, f(t) is the original signal, is the filtered signal;

[0019] S2: Obtain the optimal feature set;

[0020] S21: Calculate the correlation between the feature and the battery capacity;

[0021] Analyze the correlation between health factors and battery capacity through Pearson correlation coefficient and gray correlation coefficient. The Pearson correlation coefficient is applicable to the measurement of linear relationships, and the closer the absolute value is to 1, the stronger the correlation; the gray correlation coefficient is used to evaluate the non-linear relationship between features. These two methods can effectively identify the features closely related to the battery health state. Among them, the calculation formula of the Pearson correlation coefficient is:

[0022]

[0023] where x i is the i-th feature sequence, z is the capacity sequence, and are the average values of the i-th feature sequence and the capacity sequence respectively, is the correlation coefficient;

[0024] The calculation formula of the gray correlation degree is:

[0025]

[0026] where n is the sequence length, k represents the k-th data point, and ρ = 0.5 is the resolution coefficient;

[0027] S22: Screen out the health factors highly correlated with the battery capacity.

[0028] For each feature, if its Pearson correlation coefficient and gray correlation degree with the battery capacity > 0.5, it is considered that the feature has a strong correlation with the battery capacity;

[0029] Then, form the functional feature set f 1 , f 2 , …, f i , and sort the features according to the correlation coefficient to screen out the features with stronger correlation with the battery health state;

[0030] After screening out the features with stronger correlation, further use similarity analysis to evaluate the similarity between features. If the correlation between features > 0.9, perform redundancy removal of features to remove redundant features;

[0031] Finally, select the features highly correlated with the battery capacity and having low autocorrelation to form the optimal feature set f m , and output this feature set as the input data for the subsequent health state estimation model;

[0032] S3: Build the cell health state estimation model, specifically including:

[0033] S31: Construct model input data and initialize the model;

[0034] Based on the foregoing steps, the capacity label and the optimal feature set of the battery are obtained respectively. Each feature in the optimal feature set is normalized to ensure that the feature values are within the same numerical range, which helps to improve the training efficiency and convergence speed of the model. The normalized data will be used as the input samples of the model;

[0035] Subsequently, the training set and the test set are divided using the random slicing method. Usually, the training set accounts for 80% and the test set accounts for 20% to ensure the generalization ability of the model;

[0036] Finally, the LightGBM model is initialized and the basic parameters are set, including the depth of the tree, the learning rate, etc.;

[0037]

[0038] where xnorm is the normalized feature value; x is the initial feature set; X is the sequence of the corresponding feature;

[0039]

[0040] where is the predicted value; K is the number of decision trees; α k is the weight of each tree; h k (x) is the predicted value of the kth tree;

[0041] S32: Loss function and model optimization;

[0042] During the training process, the mean absolute error MAE is selected as the loss function because MAE is less sensitive to outliers and is more suitable for the noisy data that may appear in battery capacity prediction;

[0043] The Adam optimizer is used in the training. Its adaptive learning rate mechanism can automatically adjust the update pace of each parameter and accelerate the convergence process of the model. The Adam optimizer combines the methods of momentum and adaptive gradient and can better handle the complex non-linear relationships in the battery capacity prediction task. The optimization process can effectively fit the change trend of the battery capacity, thereby improving the prediction accuracy and stability of the model. The calculation formula of the mean absolute error is as follows:

[0044]

[0045] where MAE is the mean absolute error, n is the number of samples, is the capacity value predicted by the model, and y is the true capacity;

[0046] The operation formula of the Adam optimizer is:

[0047]

[0048] Among them, θ t+1 is the updated parameter, η is the learning rate, θ t is the current model parameter, m t is the momentum of the gradient, v t is the exponentially weighted average of the gradient square, and ∈ is a small constant to prevent division by zero;

[0049] S4: Establish a multi-stage evaluation model for the health state of the battery pack by integrating the aging condition of single cells and the inconsistency, specifically including:

[0050] S41: Extract the aging condition and inconsistency characteristics of single battery cells

[0051] The aging condition of single battery cells comes from the cell health state estimation model trained in S2. By inputting the IC health factor of each cell, the model outputs the SOH of the corresponding cell. The inconsistency characteristics of the battery pack include the voltage difference, temperature difference, average voltage, and average temperature of the cells under short-term charging. The calculation methods of the inconsistency characteristics are as follows:

[0052]

[0053] where △Vol is the voltage difference of the cell, Vol is the voltage value of the cell after short-term charging, △Temp is the temperature difference of the cell, Temp is the temperature difference of the battery after short-term charging, is the average voltage, is the average temperature, and M is the number of data samples;

[0054] S42: Obtain the capacity label and feature samples of the battery pack;

[0055] The label is the capacity of the battery pack, which can be obtained by treating the battery pack as a large battery and using the variable amperage integration method in S11. The variable amperage integration method measures the charging characteristics of the battery pack to obtain the capacity change of the entire battery pack, and then calculates the SOH label of the battery pack. The input samples of the battery pack health state evaluation model include the aging condition of the cells and the inconsistency characteristics of the battery pack. The final feature samples can be expressed as:

[0056]

[0057] where n is the number of series-connected cells in the battery pack;

[0058] S43: Initialize the model parameters and divide the data;

[0059] The model used is a long short-term memory network (LSTM) model. The LSTM model contains 2 LSTM layers and 1 fully connected layer (FCL). The feature samples are used as the input of the LSTM model. At time step t, the forget gate f of the LSTM model t and the input gate it Output gate o t and the input cell state C t are defined as follows:

[0060]

[0061] where ω is the weight matrix of each gate; U is the weight matrix of the hidden state; h t-1 represents the hidden state at time t - 1, and are four bias vectors, b is the threshold, Z i is the i-th feature sample of the input, tanh is the activation function; σ is the activation function;

[0062] To estimate the SOH of the battery pack, a fully connected layer is used to linearly transform the hidden state into the SOH output of the battery pack. The operation formula for estimating the SOH of the battery pack is:

[0063]

[0064] where h t is the temporal feature at step t, ω y and b y are the weight and threshold. During the model training process, the loss function uses MAE, and the model is optimized by minimizing the error between the predicted value and the true battery capacity. The first 20% of the aging data is selected for training in the training set, and the last 80% of the data is selected for the test set to verify the effect of the model. The Adam optimizer is used in the training to accelerate the convergence of the model in an adaptive learning rate manner, enabling it to better fit the change trend of the battery pack capacity.

[0065] (III) Beneficial effects

[0066] One of the above technical solutions has the following advantages or beneficial effects:

[0067] To address the issues that existing battery SOH evaluation methods are time-consuming, can affect the normal operation of the system, have large impacts on accuracy by factors such as environmental temperature and measurement conditions, or existing model estimation methods require a large amount of historical data support and are difficult to accurately reflect the real-time attenuation state of the battery pack, and for large-capacity energy storage systems, existing methods generally lack consideration of the consistent attenuation of the battery pack, a method for obtaining health factors and feature screening based on the IC curve is provided. It can obtain rich health factors from the IC curve, and based on the Pearson correlation coefficient method and the gray coefficient method, screen out the optimal feature set, providing a scientific basis for accurately evaluating the health state of energy storage batteries, reducing the redundancy of model inputs, improving the operating efficiency of the model, and a method for identifying the inconsistency characteristics of the battery pack is provided. It can accurately identify and quantify the inconsistency within the battery pack by analyzing characteristics such as voltage differences and temperature differences between different battery cells inside the battery pack, thereby providing a more accurate basis for evaluating the health state of the battery pack. Then, based on these two methods, a multi-stage evaluation of the health state of the energy storage battery pack is carried out, which can provide a more accurate and efficient health state prediction for large-capacity energy storage systems, and the prediction process is more interpretable, effectively improving the health management level of the energy storage system and providing reliable data support for power grid peak shaving and frequency modulation, thereby improving the operating efficiency and economic benefits of the entire energy storage system.

[0068] Other advantages, objectives, and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, objectives, and advantages of the present invention will become more obvious:

[0070] Figure 1 It is a flowchart for evaluating the health state of the energy storage battery pack;

[0071] Figure 2 It is a feature screening strategy;

[0072] Figure 3 It is a schematic diagram for modeling the cell health state evaluation model;

[0073] Figure 4 It is a schematic diagram for modeling the multi-stage evaluation model of the health state of the energy storage battery pack;

[0074] Figure 5 It is the obtained IC health factor form. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0076] With the rapid development of artificial intelligence technology, data-driven methods have shown good application prospects in the field of battery state of health prediction. This method can directly use the historical operation data of the battery to establish a prediction model without in-depth study of its complex degradation mechanism. Based on this, the present invention proposes a multi-stage evaluation method for the state of health of a new type of energy storage battery module, which conducts multi-stage evaluation of the state of health of the energy storage battery module by integrating monomer aging information and inconsistency. While ensuring the prediction accuracy, this method avoids the cumbersome battery testing process and provides a new solution for the SOH evaluation of large-scale energy storage systems;

[0077] As Figure 1 shown, the present invention mainly includes 4 steps;

[0078] Step 1: Extract the labeled capacity and health factors based on the charging data of the energy storage battery. First, through the cyclic aging experiment of battery monomers, obtain the external characteristic data of the battery during charge and discharge; then, based on the fixed SOC interval and variable ampere integration method, obtain the capacity label of the battery; finally, obtain the charging IC curve of the battery, and optimize the IC curve using Gaussian filtering. After data analysis, obtain Figure 5 the health factors in the form;

[0079] Step 2: Obtain the optimal feature set. First, calculate the Pearson correlation coefficient and gray correlation degree between the features and the battery capacity. These two methods can effectively identify the features closely related to the battery state of health; the calculation formula of the Pearson correlation coefficient is:

[0080]

[0081] where x i is the i-th feature sequence, z is the capacity sequence, and are the averages of the i-th feature sequence and the capacity sequence respectively, is the correlation coefficient, and the calculation formula of the gray correlation degree is:

[0082]

[0083] where n is the sequence length, k represents the k-th data point, and ρ = 0.5 is the resolution coefficient;

[0084] Then, screen the features with strong correlation with the battery capacity through the Pearson correlation coefficient and gray correlation degree, construct a functional feature set and sort it, and use similarity analysis to remove the redundant features with a correlation > 0.9. Finally, select the features with high correlation with the battery capacity and low autocorrelation to form the optimal feature set, providing input data for the state of health estimation model. The feature selection process is as Figure 2 shown;

[0085] Step 3: Build the cell health state estimation model. Based on the labeled capacity and the optimal feature set obtained in the previous two steps, after normalizing each feature in the optimal feature set, use the data-driven method to establish a battery capacity prediction LightGBM model. As Figure 3 shown, first, the model randomly divides the data set into a training set and a test set by inputting the normalized optimal feature set and the capacity label data, and performs modeling in an 8:2 ratio. Then, initialize the LightGBM model structure and parameters, set the training rounds and the initial learning rate of the model. Next, map the input features to the battery capacity through the fully connected layer. The model uses MAE as the loss function and continuously updates the hidden layer parameters and learning rate of the model through the Adam optimizer until the final model parameters are saved after the training cycle ends, and outputs the prediction error of the model and the parameters obtained from training;

[0086]

[0087] where n is the number of samples, is the capacity value predicted by the model, and y is the true capacity;

[0088]

[0089] where η is the learning rate, and θ t is the current model parameter, m t is the momentum of the gradient, v t is the exponentially weighted average of the squared gradient, and ∈ is a small constant to prevent division by zero;

[0090] Step 4: The process of building the multi-stage evaluation model for the health state of the energy storage battery pack is as Figure 4 shown. First, input the SOH of each cell and the inconsistency features of the battery pack. The calculation formula for the inconsistency features is as follows:

[0091]

[0092] where △Vol is the voltage difference between cells, Vol is the voltage value of the cell after short-time charging, △Temp is the temperature difference between cells, Temp is the temperature of the battery after short-time charging, is the average voltage, is the average temperature, and M is the number of data samplings;

[0093] Then, based on the aging condition of the cells and the inconsistency features of the battery pack, the feature samples of the battery pack can be expressed as:

[0094]

[0095] where n is the number of series-connected cells in the battery pack;

[0096] Next, based on the LSTM model, the temporal features of the battery pack aging process are extracted, and the fully connected layer is used to map the temporal features to the estimated value of the health state of the battery pack. At time step t, the forget gate f of the LSTM model t 、input gate i t 、output gate o t and input cell state C t are defined as follows:

[0097]

[0098] where ω is the weight matrix of each gate; U is the weight matrix of the hidden state; h t-1 represents the hidden state at time t - 1, and are four bias vectors, b is the threshold, Z i is the i-th feature sample of the input, tanh is the activation function; σ is the activation function;

[0099] To estimate the SOH of the battery pack, the fully connected layer is used to linearly transform the hidden state into the SOH output of the battery pack.

[0100]

[0101] where h t is the temporal feature at time step t, ω y and b y are the weights and thresholds. During the model training process, the loss function uses MAE, and the model is optimized by minimizing the error between the predicted value and the true battery capacity. The first 20% of the aging data is selected for training in the training set, and the last 80% of the data is selected for the test set to verify the effect of the model. The Adam optimizer is used in the training to accelerate the convergence of the model in an adaptive learning rate manner, so that it can better fit the change trend of the battery pack capacity.

[0102] The above shows and describes the basic principles, main features and advantages of the present invention. For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

[0103] In addition, it should be understood that although this specification is described in terms of embodiments, not every embodiment contains only an independent technical solution. This narrative manner of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A novel multi-stage evaluation method for the health status of energy storage battery modules, characterized in that: The following steps are involved: Step 1: Extract tag capacity and health factor based on energy storage battery charging data; First, through the cycle aging experiment of battery cells, the external characteristic data of the battery during the charging and discharging process is obtained. Then, the capacity label of the battery is obtained based on the fixed SOC interval and the variable ampere integration method; Finally, the battery charging IC curve is obtained, and Gaussian filtering is used to optimize the IC curve. After data analysis, the health factor is obtained; Step 2: Obtaining the optimal feature set; First, the correlation between health factors and battery capacity is analyzed by Pearson correlation coefficient and gray correlation coefficient, and the Pearson correlation coefficient and gray correlation degree > 0.5 are set to indicate that the feature has a strong correlation with battery capacity, thereby screening out health factors that are highly correlated with battery capacity, and then constructing and sorting the functional feature set; Then, similarity analysis was used to remove redundant features with correlation > 0.9; Finally, the features with high correlation with battery capacity and low autocorrelation are selected to form the optimal feature set, which provides input data for the health state estimation model; Step 3: Build a cell health status estimation model; First, based on the label capacity and optimal feature set obtained in steps one and two, each feature in the optimal feature set is normalized, and then a lightGBM model for battery capacity prediction is established using a data-driven approach. Then, the random slicing method is used to divide the training set and the test set, and the LightGBM model is initialized and the basic parameters are set; Next, the input features are mapped to the battery capacity through a fully connected layer; Finally, the model uses the mean absolute error as the loss function and continuously updates the model's hidden layer parameters and learning rate through the Adam optimizer until the training cycle ends, saving the final model parameters and outputting the model's prediction error and trained parameters. Step 4: Establish a multi-stage evaluation model for the health status of energy storage battery packs; First, input the SOH of each battery cell and the inconsistency characteristics of the battery pack; Then, based on the aging of the battery cells and the inconsistency characteristics of the battery pack, a characteristic sample of the battery pack is formed; Next, the time series features of the battery pack aging process are extracted based on the LSTM model, and the time series features are mapped into the estimated health status of the battery pack using a fully connected layer; Finally, in order to estimate the SOH of the battery pack, a fully connected layer is used to linearly convert the hidden state into the SOH output of the battery pack. During the model training process, the loss function uses the mean absolute error to optimize the model by minimizing the error between the predicted value and the actual battery capacity. The training set selects the first 20% of the aging data for training, and the test set selects the last 80% of the data to verify the effect of the model. The Adam optimizer is used in training to accelerate the convergence of the model with an adaptive learning rate.

2. According to claim 1, a new multi-stage evaluation method for the health status of an energy storage battery module is characterized in that: The SOC range of step 1 is 30%-80%, and the expression of the variable ampere integral method is: Among them C a is the calculated battery capacity, Δt is the sampling interval, I is the charging current, t1 and t2 are the start and end time of discharge respectively, SOC t1 is the discharge start SOC data, SOC t2 It is the SOC data at the end of discharge.

3. The multi-stage evaluation method for the health status of a novel energy storage battery module according to claim 1 is characterized in that: Step 1 The calculation formula of IC curve is as follows: Where Q is the charge, V is the voltage, I is the current, t1 is the start time of the segment charging, and t2 is the end time of the segment; Step 1 The calculation formula of Gaussian filtering is as follows: Where G(x) is the Gaussian kernel function, x is the relative position in the filter window, σ is the standard deviation of the Gaussian function, and f(t) is the original signal. is the filtered signal. The obtained health factors are shown in Table 1 in the attached figure.

4. The multi-stage evaluation method for the health status of a novel energy storage battery module according to claim 1 is characterized in that: The Pearson correlation coefficient in step 2 is suitable for measuring linear relationships. The closer the absolute value is to 1, the stronger the correlation is. The grayscale correlation coefficient is used to evaluate the nonlinear relationship between features. The Pearson correlation coefficient calculation formula is: where x i is the i-th characteristic sequence, z is the capacity sequence, and are the average values ​​of the i-th characteristic sequence and capacity sequence, respectively. is the correlation coefficient; The calculation formula of grey relational degree is: Where n is the sequence length, k represents the kth data point, and ρ=0.5 is the resolution coefficient.

5. The multi-stage evaluation method for the health status of a novel energy storage battery module according to claim 1 is characterized in that: The training set in step three accounts for 80% and the test set accounts for 20%.

6. The multi-stage evaluation method for the health status of a novel energy storage battery module according to claim 1 is characterized in that: Step 3 The calculation formula for the mean absolute error is as follows: Where MAE is the mean absolute error, n is the number of samples, is the capacity value predicted by the model, and y is the actual capacity; The calculation formula of Adam optimizer is: where θ t+1 is the updated parameter, η is the learning rate, θ t is the current model parameter, m t is the momentum of the gradient, V t is the weighted average of the squared gradients, and ∈ is a small constant to prevent division by zero.

7. The multi-stage evaluation method for the health status of a novel energy storage battery module according to claim 1 is characterized in that: Step 4 The inconsistency characteristics of the battery pack include the voltage difference, temperature difference, average voltage and average temperature of the battery cells under short-term charging. The calculation formula of the inconsistency characteristics is as follows: Where △Vol is the voltage difference of the battery cell, Vol is the voltage value of the battery cell after short-time charging, △Temp is the temperature difference of the battery cell, and Temp is the temperature difference of the battery cell after short-time charging. is the average voltage, is the average temperature, and M is the number of data samples.

8. The multi-stage evaluation method for the health status of a novel energy storage battery module according to claim 1 is characterized in that: Step 4: The characteristic samples of the battery pack are expressed as: Where n is the number of cells connected in series in the battery pack.

9. The multi-stage evaluation method for the health status of a novel energy storage battery module according to claim 1 is characterized in that: The LSTM model consists of two LSTM layers and one fully connected layer. The characteristic samples of the battery pack are used as the input of the LSTM model. When the time step is t, the forget gate f of the LSTM model t , input gate i t , output gate o t and input unit status C t The definition is as follows: Where ω is the weight matrix of each gate; U is the weight matrix of the hidden state; h t-1 represents the hidden state at time t-1, and is the four bias vectors, b is the threshold, Z i is the i-th feature sample of the input, tanh is the activation function; σ is the activation function.

10. A novel multi-stage evaluation method for the health status of an energy storage battery module according to claim 1, characterized in that: The calculation formula for estimating the battery pack SOH is: where h t is the time series feature under step length t, ω y and b y are weights and thresholds.

Images