Battery health state prediction method, system and device and storage medium

Abstract

The invention provides a battery health state prediction method, system and device and a medium, and belongs to the field of battery state estimation, and the method comprises the following steps: obtaining original electrochemical impedance spectroscopy (EIS) data of a battery; inputting the original EIS data into an automatic encoder, compressing the original EIS data into a low-dimensional vector by using the automatic encoder, and segmenting the low-dimensional vector into a plurality of time sequences; and inputting the plurality of time sequences into a domain adaptive neural network DANN, performing feature extraction on the time sequences by using the DANN to obtain domain invariant features, and predicting the health state of the battery according to the domain invariant features. According to the method, dimension reduction processing is carried out on the original EIS data through the automatic encoder, the nonlinear features in the original EIS data are extracted, the domain invariant features can be extracted in combination with the domain adaptive neural network DANN model, and the generalization ability of the model is improved.

Description

Technical Field

[0001] The present invention belongs to the field of battery state estimation, and specifically relates to a battery health state prediction method, system, device and medium. Background Art

[0002] Lithium-ion batteries have been widely used in portable devices, electric vehicles, energy storage power stations and other fields due to their advantages such as high energy density, long life and low self-discharge rate, and have become an important component of these systems. However, due to differences in working environments, lithium-ion batteries may suffer safety accidents such as thermal failure, fire and even explosion due to overcharging, high temperature and mechanical damage. At the same time, as the number of charge and discharge cycles increases, the battery itself will inevitably age, which is manifested as reduced capacity and increased internal resistance, and also increases the risk of accidents. Therefore, accurately estimating the health status of the battery is of great significance to ensure the safe and reliable operation of the battery and application system and to extend its service life.

[0003] The invention with the existing patent number CN117554822A discloses a method for predicting the battery state of health SOH based on the VAE model data enhancement algorithm, including the following steps: testing and collecting battery electrochemical impedance spectrum EIS and SOH data; using the VAE model to enhance the EIS data: using the EIS data as input data, and predicting the EIS itself through the VAE model; using the PCA model to extract the principal components of the EIS input data; constructing an MLP model to predict the battery capacity, i.e. SOH. This patent can predict the health state of the battery, however, the applicability and generalization ability of this method are limited. Since the electrochemical impedance data of different batteries have different distributions and the differences are too large, when the above model is used to process the electrochemical impedance data of different batteries, the generalization ability of the model is poor. Summary of the invention

[0004] In order to overcome the problem that the above-mentioned existing model has poor generalization ability when processing different battery electrochemical impedance data, the present invention provides a battery health state prediction method, comprising the following steps:

[0005] Obtain the original electrochemical impedance spectroscopy (EIS) data of the battery;

[0006] The original EIS data is input into an autoencoder, the autoencoder is used to compress the original EIS data into a low-dimensional vector, and the low-dimensional vector is divided into multiple time series; the multiple time series are input into a domain adaptive neural network DANN, the DANN is used to extract features of the time series to obtain domain-invariant features, and the battery health status is predicted based on the domain-invariant features.

[0007] Preferably, multiple time series are input into a domain adaptive neural network DANN, and DANN is used to extract features from the time series to obtain domain invariant features, including the following steps:

[0008] A gated recurrent unit (GRU) is used as a DANN feature extractor to extract features of the time series to obtain multiple time series features. A domain discriminator of DANN is used to determine which domain the multiple time series features belong to, and the determination result is fed back to the feature extractor. Features that are insensitive to domain information are learned based on the feedback result to obtain domain-invariant features.

[0009] Preferably, the gated recurrent unit GRU is used as a DANN feature extractor to extract features from the time series to obtain multiple time series features, specifically: the reset gate and update gate of the gated recurrent unit GRU are used to capture the dependencies in the time series, and the hidden state of the current time step is calculated based on the hidden state of the previous time step, the output of the update gate of the current time step, and the candidate hidden states to obtain multiple time series features.

[0010] Preferably, the feedback of the judgment result to the feature extractor is achieved through the gradient reversal layer of DANN.

[0011] Preferably, before inputting the multiple time series into the domain adaptive neural network DANN, the method further includes training the DANN, which specifically includes the following steps:

[0012] Obtain battery electrochemical impedance spectroscopy EIS data set;

[0013] The normalization tool is used to normalize the battery electrochemical impedance spectroscopy (EIS) data set to obtain a standardized data set;

[0014] The standardized data set is divided into a source domain data set and a target domain data set; 80% of the data in the source domain data set is used as the source domain training set, and the data in the source domain training set is used to train the feature extractor, gradient reversal layer, domain discriminator, label predictor and SOH estimator in the DANN, and the domain invariant features in the source domain training set are extracted; the first 20% of the data in the battery aging process in the target domain data set is used as the target domain training set, and the target domain training set is used to train and update the feature extractor and domain discriminator in the DANN, and the domain invariant features in the target domain training set are extracted to obtain a trained DANN model.

[0015] The present invention also provides a battery health status prediction system, comprising:

[0016] Data acquisition module, used to obtain the original electrochemical impedance spectroscopy EIS data of the battery;

[0017] A data processing module is used to input the original electrochemical impedance spectroscopy (EIS) data of the battery into an autoencoder, compress the original EIS data into a low-dimensional vector using the autoencoder, and divide the low-dimensional vector into multiple time series; input the multiple time series into a domain adaptive neural network (DANN), and use DANN to extract features from the time series to obtain domain invariant features;

[0018] A prediction module is used to predict the battery health state according to the domain invariant features.

[0019] The present invention also provides a computer device, comprising a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute the battery health status prediction method.

[0020] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the battery health status prediction method.

[0021] The battery health status prediction method, system, device and medium provided by the present invention have the following beneficial effects:

[0022] The present invention performs dimensionality reduction processing on the original EIS data through an automatic encoder, which can extract the nonlinear characteristics of the data and more complex data, thereby improving the interpretability of the model. Combined with the domain adaptive neural network DANN model, it can extract domain-invariant features in the data, making the feature representation in different domains more consistent, solving the distribution differences of different battery data, and thus improving the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiment of the present invention and its design scheme, the following briefly introduces the drawings required for this embodiment. The drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A flowchart of a method for predicting a battery health state according to an embodiment of the present invention;

[0025] Figure 2 Schematic diagram of domain adaptive neural network DANN;

[0026] Figure 3 It is the GRU structure diagram;

[0027] Figure 4These are the estimation results of different models on the 25C05 test set; (a), (b), (c) and (d) are the estimation results of AE-DANN, AE-LSTM, AE-RNN and LSTM models on the 25C05 test set, respectively. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the technical solution of the present invention and implement it, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the scope of protection of the present invention.

[0029] In the description of the present invention, it is to be understood that the terms “center”, “longitudinal”, “lateral”, “length”, “width”, “thickness”, “up”, “down”, “front”, “back”, “left”, “right”, “vertical”, “horizontal”, “top”, “bottom”, “inside”, “outside”, “axial”, “radial”, “circumferential”, etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the technical solutions of the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0030] In addition, the terms "first", "second", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance. In the description of the present invention, it should be noted that, unless otherwise clearly specified or limited, the terms "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. In the description of the present invention, unless otherwise specified, "plurality" means two or more, which will not be described in detail here.

[0031] Example

[0032] The present invention provides a method for predicting the health status of a battery, specifically Figure 1 As shown, the following steps are included:

[0033] Step 1: Obtain the original electrochemical impedance spectroscopy (EIS) data of the battery.

[0034] First, the battery data is obtained through the battery electrochemical impedance spectroscopy EIS. Figure 1 The curve graph in the upper left corner shows the change of EIS data with the degree of battery aging. The color from blue to green represents the process of battery aging.

[0035] Step 2: Input the original electrochemical impedance spectroscopy (EIS) data of the battery into the autoencoder, use the autoencoder to compress the original EIS data into a low-dimensional vector, and divide the low-dimensional vector into multiple time series; input the multiple time series into the domain adaptive neural network DANN, use DANN to extract features of the time series, obtain domain-invariant features, and predict the battery health status based on the domain-invariant features.

[0036] The automatic encoder of the present invention is an AE encoder, which consists of two parts, including an encoder and a decoder. The role of the encoder is to compress the data, which can also be described as converting the original data into other feature spaces. This conversion process is called the encoding stage. The purpose of the encoder is feature extraction, so as to obtain meaningful information to represent the data. The role of the decoder is to restore the data converted by the encoder to the original representation space, by minimizing the error between the original data and the reconstructed data obtained by decoding. If the error is as small as possible, it is considered that the feature data extracted by the encoder can represent the original data. Figure 1 The four curves in the upper right corner show the changing trend of four of the extracted latent features with the number of battery cycles. The autoencoder compresses the data features through the encoder, forms a low-dimensional representation (latent feature) at the bottleneck in the middle, and then reconstructs the data through the decoder and minimizes the reconstruction error to achieve the purpose of dimensionality reduction.

[0037] Among them, the input layer and the hidden layer constitute the encoder, and the compressed data is obtained. The encoder is:

[0038] y t =W T x t +b

[0039] In the formula, x t ={x1,x2,…,x n}∈X is the normalized data of the original EIS of the battery, W is the weight of the AE encoder, W T is the transpose of W, b is the bias parameter, y t is the output of the encoder, that is, the compressed data.

[0040] Then to reconstruct X t , map the compressed data back to the input space, define the decoder, the hidden layer and the output layer constitute the decoder, specifically:

[0041]

[0042] Where W' is the weight of the AE decoder, b' is the bias parameter of the decoder, is the output of the decoder, i.e. the reconstructed data.

[0043] Since the original data and the reconstructed EIS data are continuous, the root mean square error RMSE is introduced as the loss function of the AE autoencoder to calculate the error between the original data and the reconstructed data, and the stochastic gradient descent (SGD) can be used for optimization to minimize the error:

[0044]

[0045] In the present invention, before using the autoencoder, it is necessary to train the autoencoder, wherein the encoder is composed of two layers of fully connected layers, the first layer inputs 120 dimensions and outputs 60-dimensional data, the second layer inputs 60 dimensions and outputs 10-dimensional data, wherein the encoder outputs 10-dimensional data as the final result of dimensionality reduction. The decoder is also composed of two layers of fully connected layers, and the input and output specifications are mirror images of the encoder, that is, the first input is 10 dimensions and outputs 60-dimensional data, the second layer inputs 60 dimensions and outputs 120-dimensional data, wherein the decoder outputs 120-dimensional data as the reconstruction result. Set the loss function to RMSE, the optimizer to Adam, the batch size to 8, and the training epoch to 100.

[0046] Step 3: Input multiple time series into the domain adaptive neural network DANN, use DANN to extract features of the time series, obtain domain invariant features, and predict the battery health status based on the domain invariant features.

[0047] (1) Training DANN includes the following steps:

[0048] (a) Obtain battery electrochemical impedance spectroscopy (EIS) data set.

[0049] The data set used in the present invention is a public data set in Nature.

[0050] (b) The normalization tool is used to normalize the public data set of battery electrochemical impedance spectroscopy (EIS) to obtain a standardized data set.

[0051] First, in order to eliminate the feature importance bias caused by numerical differences, before using battery data for training and testing, it is necessary to use a standardization method to make the data fall within an agreed numerical range, and introduce a normalization tool to standardize the data:

[0052]

[0053] Among them, x' is the normalized data, x is the original data, is the mean, where σ is the standard deviation, where n represents the number of samples in the data set.

[0054] (c) The standardized dataset is divided into a source domain dataset and a target domain dataset. 80% of the data in the source domain dataset is used as the source domain training set. The data in the source domain training set is used to train the feature extractor, gradient reversal layer, domain discriminator, label predictor and SOH estimator in DANN, and the domain invariant features in the source domain training set are extracted. The first 20% of the data in the battery aging process in the target domain dataset is used as the target domain training set. The target domain training set is used to train and update the feature extractor and domain discriminator in DANN, extract the domain invariant features in the target domain training set, and obtain a battery health status prediction model.

[0055] (2) Multiple time series are input into a domain adaptive neural network (DANN), and a gated recurrent unit (GRU) is used as a DANN feature extractor to extract features from the time series to obtain domain-invariant features, and the battery health status is predicted based on the domain-invariant features.

[0056] DANN includes feature extractor, gradient reversal layer, domain discriminator, label predictor and SOH estimator, such as Figure 2 As shown in the figure. The feature extractor generates new features based on the input features, which is essentially a feature extraction structure. The label predictor predicts labels based on the features of the source domain, and it can learn the mapping relationship between the data and labels in the source domain. The domain discriminator is used to determine which domain the generated features come from. The gradient reversal layer continuously passes the inverted gradient to the feature extractor, forcing the feature extractor to learn feature representations that are insensitive to domain information and generate domain-independent features, so that the domain discriminator cannot distinguish which domain the features come from, completing the process of domain adaptation and improving the generalization ability and performance of the model in different domains.

[0057] The present invention uses the gated recurrent unit GRU as the feature extractor of the DANN network, such as Figure 3 As shown. To obtain the time series features of the data, compared with long short-term memory LSTM and recurrent neural network RNN, GRU is more concise and efficient, and can extract time series features faster. GRU has two gates, a reset gate and an update gate. The reset gate controls which past information is retained in the current time step. It determines which information is retained through the 0-1 value output by a sigmoid activation function. The closer the output value is to 1, the more information should be retained; the update gate controls which information is updated in the current time step. It also determines which information should be updated based on the value output by the sigmoid activation function. The closer the output is to 1, the more the current time step output should be used to update the hidden state. Through the reset gate and update gate, GRU can selectively retain historical information, thereby better capturing the dependencies in the time series. After calculating the outputs of the reset gate and update gate, it is necessary to further calculate the candidate hidden state of the current time step.

[0058]

[0059] Among them, W xh , W hh are weight parameters, b h is the bias parameter, the operator * represents matrix multiplication, the operator ⊙ represents matrix dot multiplication, that is, the corresponding elements in the same position are multiplied, and the candidate hidden state of the current time step is calculated through the tanh activation function. The formulas of σ and tanh activation function are as follows:

[0060]

[0061] Finally, according to the hidden state h of the previous time step t-1 and the current time step updates the gate output z t and candidate hidden states Calculate the hidden state h at the current time step t :

[0062]

[0063] For an input with a time series length of k, the input should be X = {x t ,x t+1 ,…,x t+k-1}, the corresponding GRU output hidden state should be F = {h t ,h t+1 ,…,h t+k-1}, and finally use F as the input parameter of the subsequent network, that is, Figure 1 The final output features from the green feature extractor.

[0064] For the source domain D s ={(x s ,y s )} and the target domain D t ={(x t ,y t )}, where x s and x t are the characteristics of the source domain and the target domain, which usually have different distributions, y s and t is the corresponding label, which is the battery's SOH in the battery data. In deep transfer learning, in order to enable the model to adapt to data from two domains and effectively estimate SOH, a domain discriminator is defined to determine which domain the data comes from. The domain discriminator consists of two fully connected layers and outputs the domain classification result:

[0065] domain=f(Wd F+b d )

[0066] Among them, F is the feature extracted by the feature extractor, W d is the weight parameter, b d is the bias parameter.

[0067] A gradient reversal layer is added before the domain discriminator, that is, during the training process, the model will maximize the classification error between the source domain and the target domain, that is, minimize the performance of the domain classifier, so as to encourage the feature extractor to extract features that are indistinguishable between the source domain and the target domain, that is, to extract domain-independent features for SOH prediction, so that the model trained on the source domain can have good prediction results on the target domain.

[0068] Finally, the features F extracted by the feature extractor will be used to estimate the SOH of the battery. The SOH estimator consists of two fully connected layers and can be defined as:

[0069] SOH=f(W y F+b y )

[0070] Among them, F is the feature extracted by the feature extractor, W y is the weight parameter, b y is the bias parameter.

[0071] The overall loss function in the network is:

[0072]

[0073] Where m is the number of samples, D true Denotes the true domain label, d predicted Denotes the predicted domain label.

[0074] Secondly, it is the SOH estimator loss. Since SOH is a continuous variable, RMSE is used as the loss of the SOH estimator to minimize the estimation error of SOH:

[0075]

[0076] Where n is the number of samples, SOH true Indicates the real SOH, SOH predicte represents the estimated SOH.

[0077] In order to further measure the distribution between the features output by the feature extractor and to cooperate with the model to complete domain adaptation, the MMD distance is used to measure the distribution difference. The MMD distance can make the features of the source domain and the target domain closer. Using it as a loss function can guide the feature extractor to better extract domain-independent features. The MMD distance is defined as follows:

[0078]

[0079] Among them, X s , X t Represents the data of the source domain and the target domain respectively, n s 、n t is the corresponding number of samples, Φ(·) represents a nonlinear mapping, ||·|| H is the reproducing kernel Hilbert space.

[0080] Finally, the feature extractor is used to extract the features of the source domain and target domain data, and the MMD distance is used to calculate the distance between the two distributions to obtain the loss:

[0081]

[0082] Among them, F s 、F t They represent the data of the source domain and the target domain extracted by the feature extractor respectively, k(·) represents the kernel function, and the loss function of the domain adaptive neural network DANN is:

[0083] Loss=Loss soh +Loss domain +Loss mmd

[0084] The training of the deep transfer learning model of the present invention is based on the above principles. After completing the dimensionality reduction of the EIS data, it is divided into time series for the input of the feature extractor GRU, where the length of the time series is 4, that is, the EIS data of the previous 4 cycles is used to predict the SOH of the current time step, so the input data size is (4, 10), the GRU hidden state is set to 122, the number of GRU stacking layers is 2, and the GRU is set to a bidirectional GRU, that is, the final GRU output size is (4, 244). Finally, it is expanded as the output of the feature extractor, and the extracted features are input into the SOH estimator and the domain discriminator, wherein the SOH estimator consists of a three-layer fully connected network, the first layer inputs 976 dimensions and outputs 488 dimensions of data, the second layer outputs 244 dimensions of data, and the third layer outputs 1-dimensional data as the final SOH output. The domain discriminator is similar to the SOH estimator, except that the last layer outputs 2-dimensional data to distinguish whether the data comes from the source domain or the target domain. The loss function of the SOH estimator is RMSE, the loss function of the domain discriminator is cross entropy, the optimizer is Adam, the batch size is 8, and the training is 100 epochs. DANN training includes training on the source domain and the target domain. The model outputs the features extracted by the feature extractor, the SOH estimation results, and the domain classification results. First, the data of the source domain is input to calculate the SOH estimation error and the domain classification error. Then, the target domain data is input to calculate the domain classification error. Then, the mmd distance is calculated through the features extracted from the source domain and the target domain. Finally, the error is merged and the DANN network is updated.

[0085] The standardized data set is segmented, and 25C01, 25C02, 25C03, 35C01, and 45C01 are selected as source domains for model training, and 25C05, 25C06, 35C02, and 45C02 are used as target domains for model training and testing. 20% of the data is further randomly extracted from the source domain as a validation set to verify the accuracy of SOH prediction and prevent overfitting of the model. It will not participate in the training and update of the model. The remaining 80% of the data will be used as the training set of the source domain to participate in the training of the entire model. Secondly, due to the need to extract domain-invariant features of the source and target domains, the experiment extracts the first 20% of the data in the entire battery aging process of the target domain as the target domain training set, which will only participate in the training and update of the feature extractor and domain discriminator, and will not participate in the training of the SOH predictor. Finally, 80% of the data after the cycle is used as a test set to evaluate the effectiveness of the model. During the training process, the parameters are adjusted to minimize the prediction error of SOH and maximize the error of domain classification. Finally, the accuracy of the model was evaluated by the coefficient of determination (R2), root mean square error (RMSE) and mean absolute error (MAE).

[0086] The definition of SOH battery health status is based on the capacity calculation formula:

[0087]

[0088] Among them, C i represents the battery capacity of the i-th aging cycle test, and C0 represents the rated capacity of the initial battery.

[0089] The present invention also provides a battery health status prediction system, including a data acquisition module, a data processing module and a prediction module. The data acquisition module is used to acquire the original electrochemical impedance spectrum EIS data of the battery; the data processing module is used to input the original electrochemical impedance spectrum EIS data of the battery into an autoencoder, use the autoencoder to compress the original EIS data into a low-dimensional vector, and divide the low-dimensional vector into multiple time series; input the multiple time series into a domain adaptive neural network DANN, use DANN to extract features from the time series, and obtain domain invariant features; the prediction module is used to predict the battery health status according to the domain invariant features.

[0090] The present invention also provides a computer device, including a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute a battery health status prediction method.

[0091] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program is suitable for being loaded by a processor to execute a battery health status prediction method.

[0092] In order to evaluate the effectiveness of the model in SOH prediction, MAE, RMSE and R2 are used. The mean absolute error is the average value of the error, which describes the average deviation between the true value and the estimated value. The root mean square error describes the discreteness and convergence of the estimated error and is the main reference error of this invention. The coefficient of determination is used to describe the degree of fit between the estimated value and the true value. The closer it is to 1, the better the fit of the estimated value to the true value:

[0093]

[0094] in, and They represent the real SOH and predicted SOH in the i-th cycle respectively, represents the mean of the true SOH, and

[0095] For the convenience of description, the method of the present invention is referred to as AE-DANN. Figure 4(a), (b), (c) and (d) are the estimation results of AE-DANN, AE-LSTM, AE-RNN and LSTM models on the 25C05 test set, respectively. Figure 4 As shown in the figure, the first 20% of the data is used as the target domain training set, which is the triangular curve part in the figure. This part of the data does not participate in the test. The circular curve is the true SOH value, and the cross is the estimated SOH value. Compared with other models, the AE-DANN scheme proposed in this invention has the best estimation results, R2: 0.99, RMSE: 1.075%, MAE: 0.885%.

[0096] Table 1 Evaluation indicators of the proposed model and the comparison model on the target domain dataset

[0097]

[0098] The above embodiments are only preferred specific implementation methods of the present invention, and the protection scope of the present invention is not limited thereto. Any simple changes or equivalent replacements of the technical solutions that can be obviously obtained by any technician familiar with the field within the technical scope disclosed in the present invention belong to the protection scope of the present invention.

Claims

1. A method for predicting battery health status, characterized in that: The steps include: Obtain the original electrochemical impedance spectroscopy (EIS) data of the battery; Inputting the original electrochemical impedance spectroscopy (EIS) data of the battery into an autoencoder, compressing the original EIS data into a low-dimensional vector using the autoencoder, and dividing the low-dimensional vector into multiple time series; inputting the multiple time series into a domain adaptive neural network (DANN), and using DANN to extract features from the time series to obtain domain invariant features; The battery health state is predicted based on the domain invariant features.

2. The battery health status prediction method according to claim 1, characterized in that: Inputting multiple time series into a domain adaptive neural network DANN, using DANN to extract features from the time series to obtain domain invariant features, including the following steps: A gated recurrent unit (GRU) is used as a DANN feature extractor to extract features of the time series to obtain multiple time series features. A domain discriminator of DANN is used to determine which domain the multiple time series features belong to, and the determination result is fed back to the feature extractor. Features that are insensitive to domain information are learned based on the feedback result to obtain domain-invariant features.

3. The battery health status prediction method according to claim 2, characterized in that: The gated recurrent unit GRU is used as a DANN feature extractor to extract features from the time series to obtain multiple time series features, specifically: the reset gate and update gate of the gated recurrent unit GRU are used to capture the dependency in the time series, and the hidden state of the current time step is calculated according to the hidden state of the previous time step, the output of the update gate of the current time step and the candidate hidden state to obtain multiple time series features.

4. The battery health status prediction method according to claim 2, characterized in that: The feedback of the judgment result to the feature extractor is achieved through the gradient reversal layer of DANN.

5. The battery health status prediction method according to claim 1, characterized in that: Before inputting the multiple time series into the domain adaptive neural network DANN, the DANN is trained, which specifically includes the following steps: Obtain battery electrochemical impedance spectroscopy EIS data set; The normalization tool is used to normalize the battery electrochemical impedance spectroscopy (EIS) data set to obtain a standardized data set; The standardized data set is divided into a source domain data set and a target domain data set; 80% of the data in the source domain data set is used as the source domain training set, and the data in the source domain training set is used to train the feature extractor, gradient reversal layer, domain discriminator, label predictor and SOH estimator in the DANN, and the domain invariant features in the source domain training set are extracted; the first 20% of the data in the battery aging process in the target domain data set is used as the target domain training set, and the target domain training set is used to train and update the feature extractor and domain discriminator in the DANN, and the domain invariant features in the target domain training set are extracted to obtain a trained DANN model.

6. A battery health status prediction system, characterized in that: include: Data acquisition module, used to obtain the original electrochemical impedance spectroscopy EIS data of the battery; A data processing module is used to input the original electrochemical impedance spectroscopy (EIS) data of the battery into an autoencoder, compress the original EIS data into a low-dimensional vector using the autoencoder, and divide the low-dimensional vector into multiple time series; input the multiple time series into a domain adaptive neural network (DANN), and use DANN to extract features from the time series to obtain domain invariant features; A prediction module is used to predict the battery health state according to the domain invariant features.

7. A computer device, characterized in that: It comprises a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute the battery health status prediction method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the battery health status prediction method according to any one of claims 1 to 5.

Images