Battery health state artificial intelligence estimation method based on EIS

Abstract

The invention relates to the technical field of battery health management, and provides an EIS-based battery health state artificial intelligence estimation method, which comprises the following steps of: obtaining a battery health database, the data in the battery health database comprises EIS data and battery capacity data corresponding to each charge-discharge cycle number of each battery sample in a plurality of battery samples; a TCN-Transform model is constructed, the TCN-Transform model is trained by using the battery health database, and a battery health state prediction model is obtained; obtaining current EIS data of a battery to be subjected to health state prediction, inputting the EIS data into the battery health state prediction model to obtain a battery capacity prediction value of the battery to be subjected to health state prediction, and dividing the battery capacity prediction value by a battery capacity rated value corresponding to the battery to be subjected to health state prediction to obtain a battery capacity prediction value of the battery to be subjected to health state prediction. Obtaining a battery health state estimation result; according to the method, a relatively accurate battery health state estimation result can be given.

Description

Technical Field

[0001] The present invention relates to the technical field of battery health management, and more specifically, to an artificial intelligence estimation method for battery health status based on EIS. Background Art

[0002] With the transformation of the global energy structure and the increasing severity of climate change, the demand for new energy vehicles and energy storage systems has increased dramatically. As the core energy storage and supply unit of these systems, batteries have become a key component in the development of new energy technologies. Especially in electric vehicles and large-scale energy storage equipment, the performance of batteries is directly related to the operating efficiency, service life and safety of the equipment. However, as the battery is used for a longer time, a series of aging processes will occur, resulting in a decline in its performance and may even cause safety problems such as overheating and fire. This aging phenomenon includes performance such as capacity decay, increased internal resistance, and reduced charge and discharge efficiency. The root cause is the chemical reaction and physical structure change of battery materials. Therefore, accurate prediction of the battery's state of health (SOH) and remaining service life (RUL) can ensure the safety of battery use and bring new technical improvements to the field of battery management system (BMS) optimization.

[0003] In existing technical research, there are three main types of battery health diagnosis methods: physical models, semi-empirical models, and data-driven methods. With the rise of data-driven technology, methods based on machine learning and neural networks are widely used in battery health estimation. This method does not require an in-depth understanding of the aging principle of the battery. The model can extract rules from large-scale historical data through automatic learning, reducing the complexity of manual intervention and model derivation. Selecting the right health factor has a great impact on the accuracy of the prediction. The existing technology proposes a method for battery health estimation using charge and discharge curves or incremental capacity curves combined with deep learning. However, the charge and discharge curves must fully record the voltage, current, and time data in each charge and discharge cycle, and the process is relatively cumbersome; the incremental capacity curve analysis requires high-frequency data acquisition and differential voltage calculation, which increases the burden on the battery management system; and the battery health estimation method based on the charge and discharge curve or incremental capacity curve has poor accuracy.

[0004] In summary, the existing technology has the problems of difficulty in data collection and poor accuracy. Summary of the invention

[0005] In order to overcome the defects of the above-mentioned prior art in data collection difficulty and poor accuracy, the present invention provides an EIS-based battery health status artificial intelligence estimation method with simple data collection and high accuracy.

[0006] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0007] Acquire a battery health database, wherein the data in the battery health database includes EIS data and battery capacity data corresponding to each charge and discharge cycle number of each battery sample among a plurality of battery samples;

[0008] Constructing a TCN-Transformer model, and using the battery health database to train the TCN-Transformer model to obtain a battery health status prediction model;

[0009] Obtain the current EIS data of the battery for which the health status is to be predicted, input the EIS data into the battery health status prediction model, obtain the battery capacity prediction value of the battery for which the health status is to be predicted, divide the battery capacity prediction value by the battery capacity rated value corresponding to the battery for which the health status is to be predicted, and obtain the battery health status estimation result.

[0010] The present invention also proposes a battery life prediction method based on EIS, comprising the following steps:

[0011] The number of charge and discharge cycles when the battery capacity of the battery sample is equal to the preset capacity threshold is regarded as the end-of-life of the battery sample;

[0012] Obtain the battery health database, obtain the end-of-life and current charge and discharge cycle number of several battery samples based on the battery health database, subtract the current charge and discharge cycle number from the end-of-life of the battery sample corresponding to the EIS data corresponding to the current charge and discharge cycle number, and obtain the current remaining service life corresponding to the EIS data corresponding to the current charge and discharge cycle number;

[0013] Obtain the TCN-Transformer model, and train the TCN-Transformer model using the EIS data corresponding to each charge and discharge cycle of each battery sample and the current remaining service life to obtain a remaining service life prediction model;

[0014] The current EIS data of the battery for which the remaining service life is to be predicted is obtained, and the EIS data is input into the remaining service life prediction model to obtain a predicted service life value of the battery for which the remaining service life is to be predicted.

[0015] The present invention also proposes a battery health state estimation and battery life prediction system based on EIS, which is used to implement the above-mentioned battery life prediction method based on EIS, and the system includes:

[0016] A data acquisition module is used to obtain EIS data, battery capacity data and current remaining service life corresponding to each charge and discharge cycle number of several battery samples;

[0017] A model building module, used to build the TCN-Transformer model;

[0018] A model training module, used to train the TCN-Transformer model based on the EIS data and battery capacity data corresponding to each charge and discharge cycle of the plurality of battery samples to obtain the battery health state prediction model, and to train the TCN-Transformer model based on the EIS data and current remaining service life corresponding to each charge and discharge cycle of the plurality of battery samples to obtain the remaining service life prediction model;

[0019] a battery health state estimation module, on which the battery health state prediction model is configured, for obtaining current EIS data of a battery to be predicted for health state, and inputting the obtained EIS data into the battery health state prediction model to obtain a battery capacity prediction value of the battery to be predicted for health state, and dividing the battery capacity prediction value by the battery capacity rating corresponding to the battery to be predicted for health state, to obtain a battery health state estimation result;

[0020] The remaining service life prediction module is configured with the remaining service life prediction model, which is used to obtain the current EIS data of the battery to be predicted for the remaining service life, and input the obtained EIS data into the remaining service life prediction model to obtain the service life prediction value of the battery to be predicted for the remaining service life.

[0021] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0022] The EIS (Electrochemical Impedance Spectroscopy) data is a non-invasive measurement quantity. By collecting impedance responses at different frequencies, the aging information inside the battery can be obtained. The TCN-Transformer model is trained using EIS data and battery capacity data, and the resulting battery health status prediction model can give a more accurate battery health status estimation result. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a schematic diagram of a first flow chart of the EIS-based battery health state artificial intelligence estimation method proposed in Example 1;

[0024] Figure 2 This is a schematic diagram of the structure of the TCN-Transformer model proposed in Example 1;

[0025] Figure 3 A second flow chart of the EIS-based battery health state artificial intelligence estimation method proposed in Example 2;

[0026] Figure 4 This is a schematic diagram of the battery health status estimation result proposed in Example 2;

[0027] Figure 5 This is a flow chart of the EIS-based battery life prediction method proposed in Example 3;

[0028] Figure 6 This is a schematic diagram of the battery remaining service life prediction result proposed in Example 3. DETAILED DESCRIPTION

[0029] The accompanying drawings are only used for illustrative purposes and should not be construed as limiting the present embodiment;

[0030] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;

[0031] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0032] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0033] Example 1

[0034] This embodiment proposes an artificial intelligence estimation method for battery health status based on EIS. Figure 1 This is a schematic diagram of a first flow chart of the EIS-based battery health state artificial intelligence estimation method of this embodiment;

[0035] A battery health state artificial intelligence estimation method based on EIS proposed in this embodiment includes the following steps:

[0036] S1: Acquire a battery health database, wherein the data in the battery health database includes EIS data and battery capacity data corresponding to each charge and discharge cycle number of each battery sample among a plurality of battery samples;

[0037] S2: construct a TCN-Transformer model, and use the battery health database to train the TCN-Transformer model to obtain a battery health status prediction model;

[0038] S3: Obtain the current EIS data of the battery for which the health status is to be predicted, input the EIS data into the battery health status prediction model, obtain the battery capacity prediction value of the battery for which the health status is to be predicted, divide the battery capacity prediction value by the battery capacity rated value corresponding to the battery for which the health status is to be predicted, and obtain the battery health status estimation result.

[0039] In the specific implementation process, the EIS (Electrochemical Impedance Spectroscopy) data is a non-invasive measurement quantity. By collecting impedance responses at different frequencies, the aging information inside the battery can be obtained. The TCN-Transformer model is trained using EIS data and battery capacity data, and the resulting battery health status prediction model can give a more accurate battery health status estimation result.

[0040] In an optional embodiment, the battery health database is directly obtained from the Internet or collected by itself;

[0041] The steps to collect a battery health database yourself include:

[0042] Collect several battery samples, perform a full life cycle charge and discharge cycle on each battery sample, and record the number of charge and discharge cycles, wherein the charge and discharge cycle refers to the cycle process of the battery being discharged from a fully charged state to being completely exhausted, and then charged to a fully charged state; the initial value of the number of charge and discharge cycles is 1, and the number of charge and discharge cycles is increased by one each time a charge and discharge cycle is completed, and the full life cycle refers to the charge and discharge cycle of the battery sample from a brand new state to the end of life state;

[0043] The maximum capacity of each battery sample in the discharge process under each charge and discharge cycle is counted to obtain the battery capacity of each battery sample under each charge and discharge cycle. The maximum capacity of a battery sample in the discharge process under a charge and discharge cycle is the battery capacity of the battery sample under the charge and discharge cycle.

[0044] Among them, after each battery sample is fully discharged at each charge and discharge cycle, after a preset rest time, a preset number of frequencies are selected within a preset frequency range, and an AC current of a preset number of frequencies is applied to each battery sample, and the impedance value of each battery sample at each frequency is collected to obtain EIS data of each battery sample at each charge and discharge cycle, wherein the impedance value includes a real part and an imaginary part, the real part of the impedance value is the resistance value, and the imaginary part is the reactance value.

[0045] As an exemplary illustration, when the battery capacity drops to a preset percentage of the initial capacity, the battery is considered to be at the end of its life, wherein the preset percentage can be set according to experience or experimental conditions. As an exemplary illustration, the preset percentage of this optional embodiment is selected to be 80% or 70%, that is, when the battery capacity drops to 80% or 70% of the initial capacity, the battery is considered to be at the end of its life.

[0046] In an optional embodiment, the battery samples in the battery health database and the batteries to be predicted for health status are all lithium batteries of the same type, and the types of lithium batteries include: nickel-cobalt-manganese ternary lithium batteries and lithium cobalt oxide batteries;

[0047] When the type of the lithium battery is a nickel-cobalt-manganese ternary lithium battery, each battery sample in the battery health database records its corresponding charging voltage and charging rate, that is, the data in the battery health database includes EIS data and battery capacity data corresponding to each charge and discharge cycle number of each battery sample in the plurality of battery samples at its corresponding charging voltage and charging rate;

[0048] When the type of lithium battery is a lithium cobalt oxide battery, each battery sample in the battery health database records its corresponding charging temperature, that is, the data in the battery health database includes EIS data and battery capacity data corresponding to each charge and discharge cycle number of each battery sample among the several battery samples at its corresponding charging temperature.

[0049] In an optional embodiment, the EIS data and battery capacity data in the battery health database are preprocessed respectively, and the preprocessing includes abnormal data elimination processing and normalization processing. The abnormal data elimination processing includes removing missing values, repeated values ​​and non-numeric data of the EIS data and battery capacity data, and using a polynomial interpolation method to remove abnormal noise in the EIS data and battery capacity data.

[0050] In an optional embodiment, the TCN-Transformer model includes a TCN module and a Transformer module;

[0051] The TCN module is composed of a plurality of stacked convolution blocks, wherein the input data of each convolution block is divided into two links for parallel transmission, the first link is the first causal dilated convolution layer, the first normalization layer, the first dropout layer, the second causal dilated convolution layer, the second normalization layer, the ReLU activation function layer and the second dropout layer connected in sequence; the second link is a residual connection layer, the output of the second dropout layer of the first link is added to the output of the residual connection layer of the second link as the input of the next convolution block, when the convolution block is the last convolution block, the output of the second dropout layer of the first link of the last convolution block is added to the output of the residual connection layer of the second link as the time series output by the TCN module, and the data in the time series are arranged in chronological order;

[0052] The time series output by the TCN module is input into the Transformer module;

[0053] The Transformer module includes a position encoding layer, a self-attention mechanism layer, and a fully connected layer; wherein the position encoding layer adds a position code to each position of the time series to obtain a time series with position codes; the self-attention mechanism layer is used to calculate the degree of association between each position in the time series with position codes and all other positions, and based on the degree of association, performs attention weighting on the data corresponding to each position in the time series with position codes to obtain a weighted time series; the fully connected layer is used to convert the weighted time series into a target prediction result.

[0054] As an example, Figure 2 This is a schematic diagram of the structure of the TCN-Transformer model proposed in this embodiment. Figure 2 It shows a specific structure of the TCN-Transformer model. Figure 2 Block1 represents a convolutional block of the TCN module. Figure 2 The contents of several stacked convolutional blocks are not shown.

[0055] In an optional embodiment, the position encoding expression includes:

[0056]

[0057]

[0058] Wherein, PE(pos,2i) represents the value of the 2i-th dimension of the positional encoding corresponding to the pos-th position of the time series, PE(pos,2i+1) represents the value of the 2i+1-th dimension of the positional encoding corresponding to the pos-th position of the time series, and d model Indicates the total number of dimensions of the position encoding corresponding to the pos-th position of the time series.

[0059] In an optional embodiment, when the TCN-Transformer model is trained using the battery health database, the EIS data and battery capacity data corresponding to each charge and discharge cycle of each battery sample in the battery health database are input into the TCN-Transformer model, and the TCN-Transformer model is iteratively trained until the number of iterations reaches a preset value and the determination coefficient R 2 When the maximum value is reached or the root mean square error reflection function RMSE reaches the minimum value, the iteration is stopped to obtain the battery health status prediction model;

[0060] The coefficient of determination R 2 The expressions of the root mean square error reflection function RMSE include:

[0061]

[0062]

[0063] Where Y b It indicates that the bth EIS data corresponds to the battery capacity data, and also indicates the actual result corresponding to the bth EIS data; represents the target prediction result obtained by the battery health state prediction model based on the bth EIS data, and also represents the battery capacity prediction value obtained by the battery health state prediction model based on the bth EIS data; represents the average value of all battery capacity data input; n represents the total number of EIS data input into the battery health status prediction model.

[0064] As an exemplary illustration, the objective function used for iterative training of the TCN-Transformer model may be any one of mean square error, root mean square error, mean absolute error, coefficient of determination or relative error.

[0065] As an exemplary description, this embodiment proposes a battery health state estimation system based on EIS, which is used to implement the battery health state artificial intelligence estimation method based on EIS, including:

[0066] A data acquisition module, used to obtain electrochemical impedance spectroscopy data and battery capacity data corresponding to each charge and discharge cycle number of a plurality of battery samples;

[0067] A model building module, used to build the TCN-Transformer model;

[0068] A model training module, used for training the TCN-Transformer model based on the electrochemical impedance spectroscopy data and battery capacity data corresponding to each charge and discharge cycle number of the plurality of battery samples to obtain the battery health status prediction model;

[0069] A battery health state estimation module is configured with the battery health state prediction model, which is used to obtain the current electrochemical impedance spectrum data of the battery to be predicted for health state, input the electrochemical impedance spectrum data into the battery health state prediction model, and obtain the battery capacity prediction value of the battery to be predicted for health state.

[0070] This embodiment also proposes a battery life prediction method based on EIS, which is implemented based on the above-mentioned battery health state artificial intelligence estimation method based on EIS, and includes the following steps:

[0071] S1: The number of charge and discharge cycles when the battery capacity of the battery sample is equal to the preset capacity threshold is regarded as the end life of the battery sample;

[0072] S2: Obtain the battery health database, obtain the end-of-life and current charge and discharge cycle number of several battery samples based on the battery health database, subtract the current charge and discharge cycle number from the end-of-life of the battery sample corresponding to the EIS data corresponding to the current charge and discharge cycle number, and obtain the current remaining service life corresponding to the EIS data corresponding to the current charge and discharge cycle number;

[0073] S3: Obtain the TCN-Transformer model, and train the TCN-Transformer model using the EIS data corresponding to each charge and discharge cycle of each battery sample and the current remaining service life to obtain a remaining service life prediction model;

[0074] S4: obtaining current EIS data of the battery for which remaining service life is to be predicted, inputting the EIS data into the remaining service life prediction model, and obtaining a predicted service life value of the battery for which remaining service life is to be predicted.

[0075] In an optional embodiment, when the EIS data corresponding to each charge and discharge cycle of each battery sample and the current remaining service life are used to train the TCN-Transformer model, the EIS data corresponding to each charge and discharge cycle of each battery sample and the current remaining service life are input into the TCN-Transformer model, and the TCN-Transformer model is iteratively trained until the number of iterations reaches a preset value and the determination coefficient Reaching the maximum value or root mean square error reflects the function RMSE y When the minimum value is reached, the iteration is stopped and the remaining service life prediction model is obtained;

[0076] The coefficient of determination The expressions of the root mean square error reflection function RMSE include:

[0077]

[0078]

[0079] In the formula, y c It indicates the current remaining service life corresponding to the c-th EIS data, and also indicates the actual result corresponding to the c-th EIS data; represents the target prediction result obtained by the remaining useful life prediction model based on the cth EIS data, and also represents the current remaining useful life prediction value obtained by the remaining useful life prediction model based on the cth EIS data; represents the average value of all current remaining useful lives input; N represents the total number of EIS data input into the remaining useful life prediction model.

[0080] As an exemplary illustration, the objective function used for iterative training of the TCN-Transformer model may be any one of mean square error, root mean square error, mean absolute error, coefficient of determination or relative error.

[0081] As an exemplary description, this embodiment proposes a battery life prediction system based on EIS, which is used to implement the battery life prediction method based on EIS, including:

[0082] A data acquisition module, used to obtain electrochemical impedance spectroscopy data corresponding to each charge and discharge cycle number of a plurality of battery samples and the current remaining service life;

[0083] A model building module, used to build the TCN-Transformer model;

[0084] A model training module, used for training a TCN-Transformer model based on the electrochemical impedance spectroscopy data corresponding to each charge and discharge cycle number of the plurality of battery samples and the current remaining service life to obtain the remaining service life prediction model;

[0085] The remaining service life prediction module is configured with the remaining service life prediction model, and is used to obtain the current electrochemical impedance spectrum data of the battery to be predicted for the remaining service life, input the electrochemical impedance spectrum data into the remaining service life prediction model, and obtain the service life prediction value of the battery to be predicted for the remaining service life.

[0086] As an exemplary illustration, before training the TCN-Transformer model using the EIS data corresponding to each charge and discharge cycle of each battery sample and the current remaining service life, the EIS data corresponding to each charge and discharge cycle of each battery sample and the current remaining service life are preprocessed respectively, and the preprocessing operation includes abnormal data removal and normalization, and the abnormal data removal includes removing missing values, repeated values ​​and non-numeric data of the EIS data and the current remaining service life, and using a polynomial interpolation method to remove abnormal noise in the EIS data and the current remaining service life; when training the TCN-Transformer model, the preprocessed EIS data and the current remaining service life are used to train the TCN-Transformer model; its preprocessing method is the same as the principle of the data preprocessing method of the battery health status artificial intelligence estimation method based on EIS, and the only difference is the replacement of the operation object.

[0087] It can be understood that the EIS-based battery life prediction method of this embodiment improves the EIS-based battery health status artificial intelligence estimation method of this embodiment. The options in the above-mentioned EIS-based battery health status artificial intelligence estimation method are also applicable to this method, so they will not be repeated here.

[0088] This embodiment further proposes a battery health status estimation and battery life prediction system based on EIS, which is used to implement the above-mentioned battery life prediction method based on EIS, including:

[0089] A data acquisition module is used to obtain EIS data, battery capacity data and current remaining service life corresponding to each charge and discharge cycle number of several battery samples;

[0090] A model building module, used to build the TCN-Transformer model;

[0091] A model training module, used to train the TCN-Transformer model based on the EIS data and battery capacity data corresponding to each charge and discharge cycle of the plurality of battery samples to obtain the battery health state prediction model, and to train the TCN-Transformer model based on the EIS data and current remaining service life corresponding to each charge and discharge cycle of the plurality of battery samples to obtain the remaining service life prediction model;

[0092] a battery health state estimation module, on which the battery health state prediction model is configured, for obtaining current EIS data of a battery to be predicted for health state, and inputting the obtained EIS data into the battery health state prediction model to obtain a battery capacity prediction value of the battery to be predicted for health state, and dividing the battery capacity prediction value by the battery capacity rating corresponding to the battery to be predicted for health state, to obtain a battery health state estimation result;

[0093] The remaining service life prediction module is configured with the remaining service life prediction model, which is used to obtain the current EIS data of the battery to be predicted for the remaining service life, and input the obtained EIS data into the remaining service life prediction model to obtain the service life prediction value of the battery to be predicted for the remaining service life.

[0094] Because the EIS-based battery life prediction method is implemented based on the EIS-based battery health status artificial intelligence estimation method, the EIS-based battery health status estimation and battery life prediction system includes both the contents of the EIS-based battery health status artificial intelligence estimation method and the EIS-based battery life prediction method.

[0095] Example 2

[0096] This embodiment is based on the EIS-based battery health state artificial intelligence estimation method proposed in Example 1, and proposes the following specific implementation examples:

[0097] As an exemplary description, this embodiment uses two nickel-cobalt-manganese ternary lithium batteries (NMC), namely NMC111 and NMC811, and a lithium cobalt oxide battery (LCO), namely an LCO lithium cobalt oxide button battery, to provide specific implementation examples.

[0098] Figure 3 This is a second flow chart of the EIS-based battery health state artificial intelligence estimation method proposed in this embodiment, as shown in Figure 3 As shown, the specific implementation example includes the following steps:

[0099] S1, a batch of NMC111 and NMC811 batteries are charged and discharged for the entire life cycle, and the maximum capacity during the discharge process is taken as the battery capacity. After 10-15 minutes of full charge or discharge, the electrochemical state of the battery reaches stability. The resting time is slightly different for different types of batteries and should be determined according to the actual situation of the battery. The principle is to ensure the stability of the electrochemical state. In this embodiment, the battery is allowed to rest for 15 minutes after it is fully discharged, and an alternating current of 60 different frequencies with a frequency range of approximately 0.02Hz to 20000Hz is applied to collect the impedance response (EIS). The frequency range should cover low frequency to high frequency to ensure the completeness of the battery aging information. For LCO lithium cobalt oxide button batteries, directly download the data set and save it locally.

[0100] S2, preprocess the obtained EIS data and capacity data, including removing missing values, duplicate values, non-numeric data, and processing abnormal noise in the data through polynomial interpolation to ensure the stability and reliability of the data. In order to detect data abnormality more intuitively, an image of the original data is drawn based on the preprocessed battery data, including an image of the change of EIS at different cycle numbers and an image of the attenuation of the battery capacity, so as to observe the changes in characteristic data during battery aging. After removing abnormal data, the MIN-MAX method is used to normalize EIS and capacity respectively. For example, for the normalization of EIS, the formula is satisfied:

[0101]

[0102] Among them, x i is the unnormalized impedance data, x max and x min It is the maximum and minimum impedance in different cycles at the same frequency. Normalizing the data helps the model learn the essential characteristics of battery aging, avoids errors caused by large data differences, and speeds up convergence. The normalization method of battery capacity is the same as above.

[0103] S3, divide the complete battery data set into a training set and a test set. In this embodiment, the leave-one-out method is used for cross-validation, and the data of one battery is used as the test set, and the data of the remaining batteries are used as the training set. First, a batch of NMC111 batteries are cross-validated, and then this method is applied to NMC811 batteries and LCO button batteries. For different batteries or different conditions, the principle of the model remains unchanged, and the hyperparameters such as the number of network layers are slightly adjusted.

[0104] Specifically, in this embodiment, for NMC111 batteries, one of the batteries numbered 01-08 is used as a test battery, and the EIS data of the remaining batteries are combined as a training set. When combining, ensure that the features correspond, that is, the same column represents the real part or imaginary part of the impedance at the same frequency. For example, 01-07 is selected for training, and 08 is selected for testing.

[0105] For NMC811 batteries, they are divided into three groups according to different charging conditions. The first group is numbered 01-08, with a charging voltage range of 2.5-4.3V and a charging rate of C / 40; the second group is numbered 09-11, with a charging voltage range of 2.5-4.2V and a charging rate of C / 40; the third group is numbered 12-14, with a charging voltage range of 2.5-4.2V and a charging rate of C / 20. In order to verify the applicability of the model under different charging conditions, the three groups of batteries are mixed for training to simulate more complex usage scenarios. For example, select 11 for testing and the remaining batteries for training.

[0106] The database of LCO lithium cobalt oxide button batteries is divided into three groups according to different temperature conditions. The first group is numbered 01-08, and the temperature is maintained at 25°C; the second group is numbered 09-10, and the temperature is maintained at 35°C; the third group is numbered 11-12, and the charging temperature is maintained at 45°C. In order to verify the applicability of the model under different temperature conditions, the three groups of batteries are mixed for training to simulate a usage scenario that is closer to the actual situation. For example, 01-11 is selected for training, and 12 is selected for testing.

[0107] S4, input the training set of the battery into the TCN-Transformer model. Among them, the temporal convolutional network (TCN) module processes the input EIS data by stacking multiple residual blocks, each of which contains two causal dilated convolutional layers. The convolution result passes through the dropout layer and activation layer in turn, and the information flow is promoted through jump connections and residual connections to ensure that the deep network will not be degraded due to information loss. The output sequence of the TCN module is passed to the Transformer module. The position encoding layer (Position Embedding Layer) adds position information to the input sequence and learns the position dependency in the time series. Position embedding is achieved by introducing sine and cosine functions, and the formula is as follows:

[0108]

[0109] The encoded sequence information is used to identify key patterns and features through the multi-head attention mechanism of the self-attention mechanism layer. The weight calculation formula of attention is:

[0110]

[0111] The obtained attention weights are weighted and summed with the V value, and the result is mapped to the output of the target dimension through the fully connected layer. The value obtained by the model is calculated through the regression layer for loss and back-propagation, and the weight parameters are updated to optimize the model.

[0112] Figure 4 This is a schematic diagram of the battery health status estimation result proposed in this embodiment. The model with the optimal result is applied to the battery test set to obtain the capacity prediction result, which is divided by the actual value of the initial capacity to obtain the SOH estimation value, such as Figure 4 shown. Figure 4 The three figures from left to right are the battery health status prediction results of NMC111 No. 08, NMC811 No. 11, and LCO No. 12. The horizontal axis is the number of cycles, and the vertical axis is the battery health status.

[0113] The quality of the estimation results is reflected by the coefficient of determination and the root mean square error, and the formula is:

[0114]

[0115]

[0116] Among them, Y i , and are the averages of the true value, predicted value, and true value, respectively. The closer the coefficient of determination is to 1, the better the model fit is, indicating that EIS has a strong interpretability for capacity or RUL. The smaller the root mean square error, the higher the accuracy of the model prediction.

[0117] Using the same data as input, the prediction results of TCN-Transformer in the embodiment of the present invention are compared with the prediction results of traditional neural network models such as LSTM, GPR, CNN-BiLSTM, etc., which proves that the present solution has higher accuracy, applicability and generalization ability.

[0118] Example 3

[0119] This embodiment is based on the EIS-based battery life prediction method proposed in Embodiment 1, and proposes the following specific implementation examples:

[0120] As an exemplary description, this embodiment uses two nickel-cobalt-manganese ternary lithium batteries (NMC), namely NMC111 and NMC811, and a lithium cobalt oxide battery (LCO), namely an LCO lithium cobalt oxide button battery, to provide specific implementation examples.

[0121] Figure 5 FIG. 1 is a flow chart of the EIS-based battery life prediction method proposed in this embodiment, as shown in FIG. Figure 5 As shown, the specific implementation example includes the following steps:

[0122] S1, a batch of NMC111 and NMC811 batteries are charged and discharged for the entire life cycle, and the maximum capacity during the discharge process is taken as the battery capacity. After 10-15 minutes of full charge or discharge, the electrochemical state of the battery reaches stability. The resting time is slightly different for different types of batteries and should be determined according to the actual situation of the battery. The principle is to ensure the stability of the electrochemical state. In this embodiment, the battery is allowed to rest for 15 minutes after it is fully discharged, and an alternating current of 60 different frequencies with a frequency range of approximately 0.02Hz to 20000Hz is applied to collect the impedance response (EIS). The frequency range should cover low frequency to high frequency to ensure the completeness of the battery aging information. For LCO lithium cobalt oxide button batteries, directly download the data set and save it locally.

[0123] S2, preprocess the obtained EIS data and capacity data, including removing missing values, duplicate values, non-numeric data, and processing the abnormal noise in the data through polynomial interpolation method to ensure the stability and reliability of the data. In order to detect data abnormality more intuitively, an image of the original data is drawn according to the preprocessed battery data, including the change image of EIS at different cycle numbers and the attenuation image of the battery capacity, so as to observe the changes in characteristic data during battery aging. After eliminating abnormal data, calculate 80% of the initial value of the battery capacity as the capacity corresponding to the battery's end of life, find the last cycle number in the capacity sequence that is not less than this termination capacity after data cleaning as EOL (battery end of life), and subtract the current cycle number to obtain the RUL value. Keep the EIS data of the RUL sequence length to ensure that the input and output can correspond.

[0124] Considering that in this embodiment, due to limited experimental conditions, not all batteries have reached the threshold of 80% of the initial capacity. In order to ensure the rationality of data processing, only batteries that have aged to the threshold are retained as the research of the RUL data set. Specifically, 05-08 of NMC111 batteries can be used to calculate RUL, 01-08, 10-11, 14 of NMC811 can be used to calculate RUL, and 01-03, 05-07, 09-12 of LCO button batteries can be used to calculate RUL. The EIS and RUL of the above battery serial numbers are modeled as follows:

[0125] Use the MIN-MAX method to normalize EIS and RUL respectively. For example, for the normalization of EIS, the formula is satisfied:

[0126]

[0127] Among them, x i is the unnormalized impedance data, x max and x min It is the maximum and minimum impedance in different cycles at the same frequency. Normalizing the data helps the model learn the essential characteristics of battery aging, avoids errors caused by large data differences, and speeds up convergence. The normalization method of RUL is the same as above.

[0128] S3, divide the complete battery data set into a training set and a test set. In this embodiment, the leave-one-out method is used for cross-validation, and the data of one battery is used as the test set, and the data of the remaining batteries are used as the training set. First, a batch of NMC111 batteries are cross-validated, and then this method is applied to NMC811 batteries and LCO button batteries. For different batteries or different conditions, the principle of the model remains unchanged, and the hyperparameters such as the number of network layers are slightly adjusted.

[0129] Specifically, in this embodiment, for NMC111 batteries, according to the above-mentioned EIS and RUL data processing methods, one battery in the number 05-08 is used as a test battery, and the EIS data of the remaining batteries are merged as a training set. When merging, ensure that the features correspond, that is, the same column represents the real part or imaginary part of the impedance at the same frequency. For example, 05-07 is selected for training, and 08 is selected for testing.

[0130] For NMC811 batteries, they are divided into 3 groups according to different charging conditions. According to the above EIS and RUL data processing methods, the first group retains 01-08, with a charging voltage range of 2.5-4.3V and a charging rate of C / 40; the second group retains 10-11, with a charging voltage range of 2.5-4.2V and a charging rate of C / 40; the third group retains 14, with a charging voltage range of 2.5-4.2V and a charging rate of C / 20. In order to verify the applicability of the model under different charging conditions, the three groups of batteries are mixed for training to simulate more complex usage scenarios. For example, 11 is selected for testing and the remaining batteries are trained.

[0131] For LCO button batteries, they are divided into 3 groups according to different temperature conditions. According to the above EIS and RUL data processing methods, the first group retains 01-03, 05-07, and the temperature is kept at 25°C; the second group retains 09-10, and the temperature is kept at 35°C; the third group retains 11-12, and the temperature is kept at 45°C. In order to verify the applicability of the model under different temperature conditions, the three groups of batteries are mixed for training to simulate a usage scenario that is closer to the actual situation. For example, 10 are selected for testing, and the remaining batteries are trained.

[0132] S4, input the training set of the battery into the TCN-Transformer model. Among them, the temporal convolutional network (TCN) module processes the input EIS data by stacking multiple residual blocks, each of which contains two causal dilated convolutional layers. The convolution result passes through the dropout layer and activation layer in turn, and the information flow is promoted through jump connections and residual connections to ensure that the deep network will not be degraded due to information loss. The output sequence of the TCN module is passed to the Transformer module. The position encoding layer (Position Embedding Layer) adds position information to the input sequence and learns the position dependency in the time series. Position embedding is achieved by introducing sine and cosine functions, and the formula is as follows:

[0133]

[0134] The encoded sequence information is used to identify key patterns and features through a multi-head attention mechanism. The weight calculation formula for attention is:

[0135]

[0136] The obtained attention weights are weighted and summed with the V value, and the result is mapped to the output of the target dimension through the fully connected layer. The value obtained by the model is calculated through the regression layer for loss and back-propagation, and the weight parameters are updated to optimize the model.

[0137] Figure 6 This is a schematic diagram of the battery remaining service life prediction result proposed in this embodiment. The model with the best result is applied to the battery test set to obtain the RUL prediction result, such as Figure 6 shown. Figure 6 The three figures from left to right are the remaining service life prediction results of the battery of NMC111 No. 08, the remaining service life prediction results of the battery of NMC811 No. 11, and the remaining service life prediction results of the battery of LCO No. 10. The horizontal axis is the actual remaining service life of the battery, and the vertical axis is the predicted remaining service life of the battery. The dotted line is the reference line where the predicted value and the actual value are equal.

[0138] The same or similar reference numerals correspond to the same or similar components;

[0139] The terms used to describe the positional relationship in the drawings are only used for illustrative purposes and should not be construed as limiting the present embodiment;

[0140] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. An artificial intelligence estimation method for battery health status based on EIS, characterized in that: The following steps are involved: Acquire a battery health database, wherein the data in the battery health database includes EIS data and battery capacity data corresponding to each charge and discharge cycle number of each battery sample among a plurality of battery samples; Constructing a TCN-Transformer model, and using the battery health database to train the TCN-Transformer model to obtain a battery health status prediction model; Obtain the current EIS data of the battery for which the health status is to be predicted, input the EIS data into the battery health status prediction model, obtain the battery capacity prediction value of the battery for which the health status is to be predicted, divide the battery capacity prediction value by the battery capacity rated value corresponding to the battery for which the health status is to be predicted, and obtain the battery health status estimation result.

2. The method for artificial intelligence estimation of battery health status based on EIS according to claim 1, characterized in that: The battery health database is directly obtained from the Internet or collected by itself; The steps to collect a battery health database yourself include: Collect several battery samples, perform a full life cycle charge and discharge cycle on each battery sample, and record the number of charge and discharge cycles, wherein the charge and discharge cycle refers to the cycle process of the battery being discharged from a fully charged state to being completely exhausted, and then charged to a fully charged state; the initial value of the number of charge and discharge cycles is 1, and the number of charge and discharge cycles is increased by one each time a charge and discharge cycle is completed, and the full life cycle refers to the charge and discharge cycle of the battery sample from a brand new state to the end of life state; The maximum capacity of each battery sample in the discharge process under each charge and discharge cycle is counted to obtain the battery capacity of each battery sample under each charge and discharge cycle. The maximum capacity of a battery sample in the discharge process under a charge and discharge cycle is the battery capacity of the battery sample under the charge and discharge cycle. Among them, after each battery sample is fully discharged at each charge and discharge cycle, after a preset rest time, a preset number of frequencies are selected within a preset frequency range, and an AC current of a preset number of frequencies is applied to each battery sample, and the impedance value of each battery sample at each frequency is collected to obtain EIS data of each battery sample at each charge and discharge cycle, wherein the impedance value includes a real part and an imaginary part, the real part of the impedance value is the resistance value, and the imaginary part is the reactance value.

3. The method for artificial intelligence estimation of battery health status based on EIS according to claim 1, characterized in that: The battery samples in the battery health database and the batteries to be predicted for health status are all lithium batteries of the same type, and the types of lithium batteries include: nickel-cobalt-manganese ternary lithium batteries and lithium cobalt oxide batteries; When the type of the lithium battery is a nickel-cobalt-manganese ternary lithium battery, each battery sample in the battery health database records its corresponding charging voltage and charging rate, that is, the data in the battery health database includes EIS data and battery capacity data corresponding to each charge and discharge cycle number of each battery sample in the plurality of battery samples at its corresponding charging voltage and charging rate; When the type of lithium battery is a lithium cobalt oxide battery, each battery sample in the battery health database records its corresponding charging temperature, that is, the data in the battery health database includes EIS data and battery capacity data corresponding to each charge and discharge cycle number of each battery sample among the several battery samples at its corresponding charging temperature.

4. The method for artificial intelligence estimation of battery health status based on EIS according to claim 1, characterized in that: The EIS data and battery capacity data in the battery health database are preprocessed respectively, and the preprocessing includes abnormal data elimination processing and normalization processing. The abnormal data elimination processing includes removing missing values, repeated values ​​and non-numeric data of the EIS data and the battery capacity data, and using a polynomial interpolation method to remove abnormal noise in the EIS data and the battery capacity data.

5. The method for artificial intelligence estimation of battery health status based on EIS according to claim 1, characterized in that: The TCN-Transformer model includes a TCN module and a Transformer module; The TCN module is composed of a plurality of stacked convolution blocks, wherein the input data of each convolution block is divided into two links for parallel transmission, the first link is the first causal dilated convolution layer, the first normalization layer, the first dropout layer, the second causal dilated convolution layer, the second normalization layer, the ReLU activation function layer and the second dropout layer connected in sequence; the second link is a residual connection layer, the output of the second dropout layer of the first link is added to the output of the residual connection layer of the second link as the input of the next convolution block, when the convolution block is the last convolution block, the output of the second dropout layer of the first link of the last convolution block is added to the output of the residual connection layer of the second link as the time series output by the TCN module, and the data in the time series are arranged in chronological order; The time series output by the TCN module is input into the Transformer module; The Transformer module includes a position encoding layer, a self-attention mechanism layer, and a fully connected layer; wherein the position encoding layer adds a position code to each position of the time series to obtain a time series with position codes; the self-attention mechanism layer is used to calculate the degree of association between each position in the time series with position codes and all other positions, and based on the degree of association, performs attention weighting on the data corresponding to each position in the time series with position codes to obtain a weighted time series; the fully connected layer is used to convert the weighted time series into a target prediction result.

6. The method for artificial intelligence estimation of battery health status based on EIS according to claim 5, characterized in that: The expressions of the position encoding include: Wherein, PE(pos,2i) represents the value of the 2i-th dimension of the positional encoding corresponding to the pos-th position of the time series, PE(pos,2i+1) represents the value of the 2i+1-th dimension of the positional encoding corresponding to the pos-th position of the time series, and d model Indicates the total number of dimensions of the position encoding corresponding to the pos-th position of the time series.

7. The method for artificial intelligence estimation of battery health status based on EIS according to any one of claims 1 to 6, characterized in that: When the TCN-Transformer model is trained using the battery health database, the EIS data and battery capacity data corresponding to each charge and discharge cycle of each battery sample in the battery health database are input into the TCN-Transformer model, and the TCN-Transformer model is iteratively trained until the number of iterations reaches a preset value and the determination coefficient R 2 When the maximum value is reached or the root mean square error reflection function RMSE reaches the minimum value, the iteration is stopped to obtain the battery health status prediction model; The coefficient of determination R 2 The expressions of the root mean square error reflection function RMSE include: Where Y b It indicates that the bth EIS data corresponds to the battery capacity data, and also indicates the actual result corresponding to the bth EIS data; represents the target prediction result obtained by the battery health state prediction model based on the bth EIS data, and also represents the battery capacity prediction value obtained by the battery health state prediction model based on the bth EIS data; represents the average value of all battery capacity data input; n represents the total number of EIS data input into the battery health status prediction model.

8. A battery life prediction method based on EIS, implemented based on the battery health state artificial intelligence estimation method based on EIS according to any one of claims 1 to 7, characterized in that: The following steps are involved: The number of charge and discharge cycles when the battery capacity of the battery sample is equal to the preset capacity threshold is regarded as the end-of-life of the battery sample; Obtain the battery health database, obtain the end-of-life and current charge and discharge cycle number of several battery samples based on the battery health database, subtract the current charge and discharge cycle number from the end-of-life of the battery sample corresponding to the EIS data corresponding to the current charge and discharge cycle number, and obtain the current remaining service life corresponding to the EIS data corresponding to the current charge and discharge cycle number; Obtain the TCN-Transformer model, and train the TCN-Transformer model using the EIS data corresponding to each charge and discharge cycle of each battery sample and the current remaining service life to obtain a remaining service life prediction model; The current EIS data of the battery for which the remaining service life is to be predicted is obtained, and the EIS data is input into the remaining service life prediction model to obtain a predicted service life value of the battery for which the remaining service life is to be predicted.

9. The method for predicting battery life based on EIS according to claim 8, characterized in that: When the EIS data corresponding to each charge and discharge cycle of each battery sample and the current remaining service life are used to train the TCN-Transformer model, the EIS data corresponding to each charge and discharge cycle of each battery sample and the current remaining service life are input into the TCN-Transformer model, and the TCN-Transformer model is iteratively trained until the number of iterations reaches a preset value and the determination coefficient Reaching the maximum value or root mean square error reflects the function RMSE y When the minimum value is reached, the iteration is stopped and the remaining service life prediction model is obtained; The coefficient of determination The expressions of the root mean square error reflection function RMSE include: In the formula, y c It indicates the current remaining service life corresponding to the c-th EIS data, and also indicates the actual result corresponding to the c-th EIS data; represents the target prediction result obtained by the remaining useful life prediction model based on the cth EIS data, and also represents the current remaining useful life prediction value obtained by the remaining useful life prediction model based on the cth EIS data; represents the average value of all current remaining useful lives input; N represents the total number of EIS data input into the remaining useful life prediction model.

10. A battery health status estimation and battery life prediction system based on EIS, used to implement the battery life prediction method based on EIS according to claim 8 or 9, characterized in that: include: A data acquisition module is used to obtain EIS data, battery capacity data and current remaining service life corresponding to each charge and discharge cycle number of several battery samples; A model building module, used to build the TCN-Transformer model; A model training module, used to train the TCN-Transformer model based on the EIS data and battery capacity data corresponding to each charge and discharge cycle of the plurality of battery samples to obtain the battery health state prediction model, and to train the TCN-Transformer model based on the EIS data and current remaining service life corresponding to each charge and discharge cycle of the plurality of battery samples to obtain the remaining service life prediction model; a battery health state estimation module, on which the battery health state prediction model is configured, for obtaining current EIS data of a battery to be predicted for health state, and inputting the obtained EIS data into the battery health state prediction model to obtain a battery capacity prediction value of the battery to be predicted for health state, and dividing the battery capacity prediction value by the battery capacity rating corresponding to the battery to be predicted for health state, to obtain a battery health state estimation result; The remaining service life prediction module is configured with the remaining service life prediction model, which is used to obtain the current EIS data of the battery to be predicted for the remaining service life, and input the obtained EIS data into the remaining service life prediction model to obtain the service life prediction value of the battery to be predicted for the remaining service life.

Images