Deep learning method and device for cell failure prediction in BMS

Abstract

The invention relates to the technical field of battery monitoring, in particular to a deep learning method and device for battery cell failure prediction in a BMS, and the method comprises the steps: collecting the multi-mode operation data of a battery cell, and dividing the multi-mode operation data into time sequence samples; time domain features, frequency domain features and differential features of the time sequence samples are extracted, feature selection is carried out by using Pearson correlation coefficients and mutual information, and dimensionality reduction is carried out through principal component analysis to construct a data set; a cell failure prediction model is constructed based on the gating circulation unit model, and the model performance is optimized in combination with an attention mechanism; on the basis of predicting the failure of the battery cell, predicting the health state and failure time of the battery, introducing a self-adaptive failure threshold mechanism, and dynamically adjusting a failure judgment standard; and constructing an incremental learning framework based on sample selection, and dynamically adjusting the prediction model through a new working condition sample. According to the method, the accuracy and robustness of cell failure prediction and the adaptability to complex working conditions can be remarkably improved, and more reliable decision support is provided for a battery management system.

Description

Technical Field

[0001] The present application relates to the field of battery monitoring technology, and specifically to a deep learning method and device for predicting battery cell failure in a BMS. Background Art

[0002] The battery management system (BMS) is the core guarantee for the operation of the battery pack. As the basic unit of the lithium battery, the operating status of the battery cell directly affects the performance, safety and life of the entire battery pack.

[0003] Traditional battery cell failure prediction methods are mostly based on empirical models or simple physical models. These methods usually rely on limited prior knowledge and are difficult to fully capture the complex nonlinear dynamic characteristics of battery cell operation. In terms of data processing, existing technologies have low efficiency in utilizing multimodal battery cell operation data. Although modern battery cells can collect a variety of operating parameters (such as voltage, current, temperature, etc.) through sensors, existing methods usually lack the ability to effectively fuse and deeply analyze these high-dimensional data, which makes it difficult for the model to fully utilize data information and affects the prediction effect.

[0004] Most existing cell failure prediction methods lack the ability to adapt to changes in operating conditions. The operating environment and conditions of cells in different application scenarios may vary significantly, and fixed models are difficult to cope with such dynamic changes, resulting in unstable or inaccurate prediction results.

[0005] In summary, the existing technology has obvious deficiencies in terms of accuracy of battery cell failure prediction, multimodal data processing capabilities, adaptability to working conditions, and prediction range, and more advanced technical means are urgently needed to solve the above problems.

[0006] In view of this, the present application proposes a deep learning method and device for predicting battery cell failure in BMS. Summary of the invention

[0007] To achieve the above objectives, the present application provides a deep learning method and device for predicting cell failure in a BMS. The specific technical solution is as follows: Deep learning methods for predicting cell failure in BMS include: Collect multi-modal battery cell operation data, pre-process the collected multi-modal battery cell operation data, and divide it into time series samples using a sliding window method; Extracting time domain features, frequency domain features and differential features of the time series samples, performing feature selection using the Pearson correlation coefficient and mutual information, and performing dimensionality reduction processing using the principal component analysis method to construct a data set; Build a cell failure prediction model based on the gated recurrent unit (GRU) model, train the GRU model using data from the dataset, add an attention mechanism to the GRU model, train the GRU model, and optimize the model parameters; The cell failure prediction model is expanded to predict the battery health status and failure time while predicting cell failure, and an adaptive failure threshold mechanism is introduced; In response to the new working conditions generated when the battery cell is working, an incremental learning framework based on sample selection is constructed, and the battery cell failure prediction model is adjusted through the new working condition samples.

[0008] Preferably, by installing multiple sensors on the battery cells, multi-modal battery cell operation data during the battery cell operation process is collected, including voltage, current, temperature and pressure; The collected multimodal battery cell operation data is preprocessed, including outlier detection and removal; the preprocessed data is time aligned and the time-aligned data is normalized; the normalized multimodal battery cell operation data is divided into time series samples using a sliding window approach.

[0009] Preferably, for each time series sample, feature extraction and calculation are performed, including: extracting time domain features, extracting frequency domain features, and calculating differential features; The Pearson correlation coefficient and mutual information are used to perform feature selection on the extracted and calculated time domain features, frequency domain features, and differential features; The selected features are subjected to principal component analysis and dimensionality reduction to obtain a feature matrix; the reduced feature matrix is ​​combined with the corresponding label vector to form a data set.

[0010] Preferably, a cell failure prediction model based on a gated recurrent unit (GRU) model is constructed, wherein the input layer of the GRU model receives a feature vector of a time series sample, and the output layer outputs the cell failure probability; between the input layer and the output layer of the GRU model, a plurality of GRU hidden layers are arranged, and each hidden layer is arranged with a plurality of GRU units; An attention mechanism is introduced into the constructed GRU model, so that the GRU model can focus on the time series sample information; the cross-validation method is used to tune the hyperparameters of the GRU model.

[0011] Preferably, while predicting the failure of the battery cell, two auxiliary tasks are introduced, namely, battery health state prediction and failure time prediction; the battery health state is defined as the ratio of the current capacity of the battery to the initial capacity, and the failure time is defined as the remaining number of cycles of the battery from the current moment to the moment of failure; the battery health state prediction and failure time prediction are modeled as regression problems.

[0012] Preferably, based on the constructed GRU model, output branches for battery health status prediction and failure time prediction are introduced, and feature representations of time series samples are extracted through the GRU encoder in the GRU model, and then mapped to outputs of different tasks through a fully connected layer; An adaptive failure threshold mechanism is introduced to dynamically adjust the cell failure probability threshold according to the battery health status prediction value and failure time prediction value.

[0013] Preferably, an incremental learning framework based on sample selection is constructed, and parameters of the pre-trained cell failure prediction model are adjusted by selecting new working condition samples; the selection strategies of the new working condition samples include random selection and diversity selection; An incremental learning objective function is constructed, and the updated model parameters are obtained by balancing the new operating condition sample fitting and the parameter update amplitude. The incremental learning objective function is optimized using a small batch gradient descent algorithm to obtain a cell failure prediction model that is adapted to the new operating conditions.

[0014] A deep learning device for predicting cell failure in a BMS, which is used to implement the deep learning method for predicting cell failure in a BMS, comprising: a data acquisition module, a feature extraction module, a GRU model module, a function expansion module, and an incremental learning module; The data acquisition module is used to collect multi-modal battery cell operation data, pre-process the collected multi-modal battery cell operation data, and divide it into time series samples in a sliding window manner; The feature extraction module is used to extract the time domain features, frequency domain features and differential features of the time series samples, perform feature selection using the Pearson correlation coefficient and mutual information, and perform dimensionality reduction processing using the principal component analysis method to construct a data set; The GRU model module is used to build a cell failure prediction model based on the gated recurrent unit (GRU) model, train the GRU model through data set data, add an attention mechanism to the GRU model, train the GRU model, and optimize model parameters; The function extension module is used to expand the battery cell failure prediction model, predict the battery health status and failure time while predicting the battery cell failure, and introduce an adaptive failure threshold mechanism; The incremental learning module constructs an incremental learning framework based on sample selection for new operating conditions generated when the battery cell is working, and adjusts the battery cell failure prediction model through new operating condition samples.

[0015] An electronic device comprises: a processor and a memory, wherein the memory stores a computer program that can be called by the processor; the processor executes the deep learning method for predicting battery cell failure in the BMS by calling the computer program stored in the memory.

[0016] A computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the deep learning method for predicting cell failure in a BMS.

[0017] Beneficial effects of the present application: The present application comprehensively captures the dynamic changes of the operating status of the battery cell through multimodal data acquisition and sliding window division, thereby improving the temporal integrity of the sample and the accuracy of the prediction.

[0018] This application extracts multiple features and performs selection and dimensionality reduction to reduce redundant information, improve data quality, reduce model complexity, and enhance the model's sensitivity to key features.

[0019] This application constructs and trains a GRU model that introduces an attention mechanism to effectively capture the long-term dependencies of time series data and improve the accuracy of battery cell failure prediction and the generalization ability of the model.

[0020] This application expands the model function based on the constructed GRU model, predicts SOH and failure time at the same time, introduces an adaptive failure threshold mechanism, and improves the applicability and prediction reliability of the model in multi-task scenarios.

[0021] This application adapts to new working condition samples through an incremental learning framework, improves the adaptability and dynamic adjustment capabilities of the model, and realizes real-time optimization of the battery cell failure prediction model.

[0022] The method of the present application constructs a highly accurate and adaptable battery cell failure prediction system through multimodal data collection, feature extraction, attention mechanism optimization, task expansion and incremental learning framework, providing more intelligent and comprehensive monitoring and prediction capabilities for the battery management system (BMS). BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flow chart of the deep learning method for predicting cell failure in the BMS provided in this application; Figure 2 A flow chart of constructing a cell failure prediction model for a deep learning method for cell failure prediction in a BMS provided in this application; Figure 3 A flowchart of the incremental learning framework for constructing a deep learning method for predicting cell failure in a BMS provided in this application; Figure 4 This is a structural diagram of the deep learning device for predicting battery cell failure in the BMS provided in this application. DETAILED DESCRIPTION

[0024] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are described in detail below in conjunction with the drawings in the specification.

[0025] In the following description, many specific details are set forth to facilitate a full understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without violating the connotation of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0026] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present application. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0027] Example 1

[0028] Reference Figures 1 to 3 The first embodiment of the present application provides a deep learning method for predicting battery cell failure in a BMS.

[0029] Step 1: Collect multi-modal battery cell operation data, pre-process the collected multi-modal battery cell operation data, and divide it into time series samples using a sliding window method.

[0030] Exemplarily, by installing multiple sensors on the battery cells, multimodal battery cell operation data including voltage, current, temperature and pressure are collected during the battery cell operation process; wherein, the voltage and current data are collected by sensors with an accuracy of 1mV and 1mA, and the sampling frequency is 100Hz; the temperature data is collected by a thermocouple with an accuracy of 1°C, and the sampling frequency is 1Hz; the pressure data is collected by a pressure sensor with an accuracy of 1kPa, and the sampling frequency is 1Hz.

[0031] The collected multi-modal battery cell operation data is used to detect and remove outliers; an outlier detection method based on the interquartile range (IQR) is used to detect and remove outliers in the data. , calculate its difference with the first quartile The difference Sum and third quartile The difference : ; like or , then are marked as outliers and removed; .

[0032] Time alignment: Since the data sampling frequencies of different modes are different, the data needs to be time aligned. Based on the voltage and current data, the temperature and pressure data are linearly interpolated to align them with the voltage and current data in the time dimension.

[0033] Assume the timestamps of the voltage and current data are , the timestamp of the temperature data is , for any time , find the two temperature data timestamps closest to it and , calculated by linear interpolation Temperature value at the moment : ; in, and Separately for the moment and The time alignment method for pressure data is the same as that for temperature data.

[0034] The multi-modal battery cell operation data after time alignment is numerically normalized and the data of each mode is mapped to In the range of; using the maximum and minimum value normalization method, for each mode of data , calculate its maximum value and minimum value , normalized to: ;in, is the normalized data value.

[0035] The normalized multimodal cell operation data is divided into time series samples by using a sliding window method. The window size is set to , the sliding step length is , then The time series samples are: ; in, For the The multimodal data vector of time series samples is the time series sample; the label of the time series sample is the failure state of the battery cell after the last time step of the sample, and the time series sample Tags Used to mark the failure of the battery after the sample is completed. time steps and then fail, then ,otherwise .

[0036] Step 2: Extract the time domain features, frequency domain features and differential features of the time series samples, use the Pearson correlation coefficient and mutual information to select features, and use the principal component analysis method to reduce the dimension and construct a data set.

[0037] For each time series sample , extract time domain features, including: mean , Standard Deviation , RMS , Peak , Peak Factor , skewness Kurtosis .

[0038] Extract for each time series sample , use the fast Fourier transform (FFT) to convert it to the frequency domain and get the frequency domain signal ;from Extract frequency domain features, including: frequency domain energy , frequency domain entropy , frequency domain peak And the frequency domain center of gravity .

[0039] For each time series sample , calculate its first-order difference sequence : ;from Extract time domain statistical features as differential features. Indicates The time series samples are time steps of multimodal data vector.

[0040] The Pearson correlation coefficient and mutual information are used to perform feature selection on the extracted time domain features, frequency domain features and difference features.

[0041] For each feature , calculate its difference with the label Pearson correlation coefficient : ; in, Indicates Time series samples The value of the feature, and Respectively represent characteristics and tags The mean of .

[0042] For each feature , calculate its difference with the label Mutual information : ;

[0043] in, Representation characteristics and tags The joint probability distribution of and Respectively represent characteristics and tags The marginal probability distribution of .

[0044] According to the size of the Pearson correlation coefficient and mutual information, the top Features Perform principal component analysis (PCA) dimensionality reduction on the selected features; suppose the original feature matrix is ,in is the number of time series samples, is a sparse matrix; After centralization and standardization, we get the matrix ,right Perform eigenvalue decomposition: ;in, is the eigenvector matrix, is the transpose of the eigenvector matrix, is a diagonal matrix with eigenvalues ​​on the diagonal elements.

[0045] Arrange the eigenvalues ​​and select the first feature vectors, forming a matrix , the original feature matrix Mapping to a low-dimensional space: ;in, is the feature matrix after dimensionality reduction.

[0046] The feature matrix after dimension reduction The corresponding label vector Combine to form a complete data set ,in For the The feature vector of time series samples, For the The labels of the time series samples.

[0047] Step 3: Build a cell failure prediction model based on the gated recurrent unit (GRU) model, train the GRU model using data from the dataset, add an attention mechanism to the GRU model, train the GRU model, optimize the model parameters, and perform hyperparameter tuning on the GRU model.

[0048] The input layer of the gated recurrent unit (GRU) model receives the feature vector of the time series sample, and the output layer outputs the probability of battery cell failure. Between the input layer and the output layer, multiple GRU hidden layers are set, and each hidden layer contains multiple GRU units.

[0049] For The input vector of time steps , the update formula of the GRU unit is as follows: ; ; ; ; in, To reset the gate, To update the gate, is a candidate hidden state, is the current hidden state, is the weight matrix, is the bias vector, is the Sigmoid activation function, is the hyperbolic tangent activation function, is element-wise multiplication.

[0050] Based on the GRU model, the attention mechanism is introduced to enable the GRU model to adaptively focus on the key information in the time series samples. The hidden state of time steps , calculate its attention weight with all time-step hidden states: ; ; ; in, Indicates The time step is The attention score of time steps, is the learnable parameter of the attention mechanism, Indicates The hidden state vector of time steps, is the bias vector, For the The time step is The attention weights for time steps, For the The context vector of time steps, is the total number of time steps, Represents an exponential function with a natural constant as its base.

[0051] The context vector With the current hidden state Concatenation as the final feature representation: ;in, Represents a vector concatenation operation.

[0052] The feature representation of the last time step is Pass it into the fully connected layer and use the Sigmoid activation function to get the cell failure probability: ;in, is the weight matrix of the output layer, is the bias vector.

[0053] The mean square error (MSE) is used as the loss function to measure the difference between the predicted value and the true value.

[0054] The Adam optimization algorithm is used to train the model, adaptively adjust the learning rate, and accelerate convergence; the update formula of the Adam optimization algorithm is as follows: ; ; ; ; ; in, For the The first moment estimate of the gradient at time steps, For the The first moment estimate of the gradient at time steps, For the The second-order moment estimate of the gradient at time steps, For the The second-order moment estimate of the gradient at time steps, is the modified value of the first moment estimate of the gradient, is the modified estimate of the second moment of the gradient, is the attenuation rate, For the The gradient of time steps, is the learning rate, is the smoothing term, For the The model parameters for time steps, For the The model parameters for each time step.

[0055] The cross-validation method is used to tune the hyperparameters of the model, including: the number of hidden layers of the GRU model, the number of units in each layer, the dimension of the attention mechanism, the number of fully connected layers, the number of units in each layer, the learning rate, the batch size, and the regularization coefficient.

[0056] Through hyperparameter tuning, the model parameter configuration with optimal performance is obtained for subsequent prediction tasks.

[0057] Step 4: Expand the cell failure prediction model into a multi-task learning framework to predict the battery state of health (SOH) and failure time while predicting cell failure, and introduce an adaptive failure threshold mechanism.

[0058] While predicting cell failure, battery state of health (SOH) prediction and failure time prediction are introduced.

[0059] Battery State of Health (SOH) prediction: The battery state of health SOH is defined as the ratio of the current capacity of the battery to the initial capacity, with a value range of [0, 1]. The battery state of health SOH prediction model is a regression problem, and the loss function is the mean square error MSE: ; in, is the mean square error MSE value, For the The battery health status SOH prediction value of time series samples, For the The actual battery health status SOH value of the time series samples.

[0060] Failure time prediction: The failure time is defined as the number of remaining cycles of the battery from the current moment to the failure moment, with a value range of [0, +∞); the failure time prediction model is modeled as a regression problem, and the loss function is the mean absolute error MAE: ; in, Mean absolute error MAE value, For the The failure time prediction value of time series samples, For the The actual failure time value of the time series samples.

[0061] Based on the constructed GRU model, two output branches of auxiliary tasks are introduced into the GRU model; the feature representation of time series samples is extracted through the GRU encoder in the GRU model, and then mapped to the output of different tasks through the task-specific fully connected layer.

[0062] The battery health status SOH prediction branch in the GRU model is expressed as: ; The failure time prediction branch in the GRU model is expressed as: ;in, is the hidden state of the last time step of the GRU encoder, are the weight vectors of the battery health status SOH and failure time prediction branches, is the bias term, They are the battery health status prediction value and failure time prediction value respectively.

[0063] Joint loss function for: ;in, is the cross entropy loss function for the cell failure prediction task, The weight hyperparameter for balancing the losses of the three tasks.

[0064] An adaptive failure threshold mechanism is introduced to dynamically adjust the cell failure probability threshold according to the battery health status prediction value and failure time prediction value.

[0065] Defining the cell failure probability threshold for: ;in, is the weight vector for threshold calculation, is the bias term, It is the Sigmoid activation function, which maps the threshold to the range of (0, 1).

[0066] In order to encourage the GRU model to learn a suitable failure probability threshold, a threshold penalty term is introduced into the loss function: ;in, is the preset target threshold, A weight hyperparameter to balance the threshold penalty term with other task losses.

[0067] Add the threshold penalty term to the joint loss function: ; Threshold judgment In the prediction stage, the cell failure probability is compared with the dynamic threshold to obtain the final failure prediction result: ; in, is the predicted cell failure probability result, is the cell failure probability.

[0068] Through the multi-task learning framework and adaptive failure threshold mechanism, the SOH and failure time information can be fully utilized to improve the accuracy and interpretability of battery cell failure prediction.

[0069] Step 5: In response to the new working conditions generated when the battery cell is working, an incremental learning framework based on sample selection is constructed, and the battery cell failure prediction model is adjusted through the new working condition samples.

[0070] When new working conditions are generated during the operation of the battery cell, representative samples are selected from the data collected under the new working conditions to form a new working condition sample set. ,in For the The feature vector of a new working condition sample, For the corresponding label.

[0071] Randomly select from new operating data samples constitute a new working condition sample set; Diversity selection: select samples that are significantly different from existing training samples samples, a metric such as Euclidean distance or cosine similarity can be used.

[0072] Assume that the parameters of the pre-trained model are The goal of incremental learning is to fine-tune the model parameters on new working condition samples to obtain updated model parameters , the loss function of incremental learning is: ; in, is the loss function of incremental learning, is the number of the new working condition sample, , For the model, input The predicted output on is the loss function for the incremental learning task (such as cross entropy loss or mean square error loss), is the regularization coefficient, which is used to balance the amplitude of new sample fitting and parameter updating.

[0073] The optimization goal of the incremental learning objective function is: , Indicates taking The value with the smallest value.

[0074] Use the mini-batch gradient descent algorithm to optimize the incremental learning objective function. Input: pre-trained model parameters ,New working condition sample set , learning rate , regularization coefficient , small batch size ,Iterations ; Output: Updated model parameters .

[0075] initialization: , ,from Random selection Samples constitute a small batch of samples , calculate the loss function on the mini-batch : ; Calculate the gradient of the loss function: ; Represents the loss function for the parameter The gradient of Update model parameters: , , Indicates the updated parameter value of the current iteration step, The parameter value of the previous iteration step, Represents the final iteration step The parameter value after.

[0076] Through the incremental learning framework, the pre-trained battery cell failure prediction model can be effectively adapted to new operating conditions, thereby improving the generalization performance and practical application value of the model.

[0077] Example 2

[0078] Reference Figure 4 The second embodiment of the present application provides a deep learning device for predicting battery cell failure in a BMS, including: a data acquisition module, a feature extraction module, a GRU model module, a function expansion module and an incremental learning module.

[0079] The data acquisition module is used to collect multi-modal battery cell operation data, pre-process the collected multi-modal battery cell operation data, and divide it into time series samples in a sliding window manner.

[0080] The feature extraction module is used to extract the time domain features, frequency domain features and differential features of the time series samples, use the Pearson correlation coefficient and mutual information to perform feature selection, and use the principal component analysis method to perform dimensionality reduction processing to construct a data set.

[0081] The GRU model module is used to build a cell failure prediction model based on the gated recurrent unit (GRU) model, train the GRU model through data set data, add an attention mechanism to the GRU model, train the GRU model, and optimize model parameters.

[0082] The function extension module is used to expand the cell failure prediction model, predict the battery health state (SOH) and failure time while predicting cell failure, and introduce an adaptive failure threshold mechanism.

[0083] The incremental learning module constructs an incremental learning framework based on sample selection for new operating conditions generated when the battery cell is working, and adjusts the battery cell failure prediction model through new operating condition samples.

[0084] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also change, modify, replace and modify the above-mentioned embodiments without departing from the scope of protection of the purpose of the present application and the claims, which are all within the protection of the present application.

Claims

1. A deep learning method for predicting cell failure in BMS, characterized by: include: Collect multi-modal battery cell operation data, pre-process the collected multi-modal battery cell operation data, and divide it into time series samples using a sliding window method; Extracting time domain features, frequency domain features and differential features of the time series samples, performing feature selection using the Pearson correlation coefficient and mutual information, and performing dimensionality reduction processing using the principal component analysis method to construct a data set; Build a cell failure prediction model based on the gated recurrent unit (GRU) model, train the GRU model using data from the dataset, add an attention mechanism to the GRU model, train the GRU model, and optimize the model parameters; The cell failure prediction model is expanded to predict the battery health status and failure time while predicting cell failure, and an adaptive failure threshold mechanism is introduced; In response to the new working conditions generated when the battery cell is working, an incremental learning framework based on sample selection is constructed, and the battery cell failure prediction model is adjusted through the new working condition samples.

2. The deep learning method for predicting cell failure in BMS according to claim 1, characterized in that: By installing multiple sensors on the battery cells, multimodal battery cell operation data is collected during the battery cell operation process, including voltage, current, temperature and pressure; the collected multimodal battery cell operation data is preprocessed, including outlier detection and removal; the preprocessed data is time-aligned and the time-aligned data is normalized; and the normalized multimodal battery cell operation data is divided into time series samples using a sliding window method.

3. The deep learning method for predicting cell failure in BMS according to claim 2, characterized in that: For each time series sample, feature extraction and calculation are performed, including: extracting time domain features, extracting frequency domain features, and calculating differential features; using the Pearson correlation coefficient and mutual information to perform feature selection on the extracted and calculated time domain features, frequency domain features, and differential features; The selected features are subjected to principal component analysis and dimensionality reduction to obtain a feature matrix; the reduced feature matrix is ​​combined with the corresponding label vector to form a data set.

4. The deep learning method for predicting cell failure in BMS according to claim 3, characterized in that: A cell failure prediction model based on the gated recurrent unit (GRU) model is constructed. The input layer of the GRU model receives the feature vector of the time series sample, and the output layer outputs the cell failure probability. Multiple GRU hidden layers are set between the input layer and the output layer of the GRU model, and multiple GRU units are set in each hidden layer. An attention mechanism is introduced into the constructed GRU model, so that the GRU model can focus on the time series sample information; the cross-validation method is used to tune the hyperparameters of the GRU model.

5. The deep learning method for predicting cell failure in BMS according to claim 4, characterized in that: While predicting cell failure, two auxiliary tasks are introduced, namely battery health status prediction and failure time prediction; The battery health status is defined as the ratio of the current capacity of the battery to the initial capacity, and the failure time is defined as the number of remaining cycles of the battery from the current time to the failure time; Battery health status prediction and failure time prediction are modeled as regression problems.

6. The deep learning method for predicting cell failure in BMS according to claim 5, characterized in that: Based on the constructed GRU model, the output branches of battery health status prediction and failure time prediction are introduced. The feature representation of time series samples is extracted through the GRU encoder in the GRU model, and then mapped to the output of different tasks through the fully connected layer. An adaptive failure threshold mechanism is introduced to dynamically adjust the cell failure probability threshold according to the battery health status prediction value and failure time prediction value.

7. The deep learning method for predicting cell failure in a BMS according to claim 6, characterized in that: An incremental learning framework based on sample selection is constructed. By selecting new operating condition samples, the parameters of the pre-trained battery cell failure prediction model are adjusted. The selection strategies for new operating condition samples include random selection and diversity selection. An incremental learning objective function is constructed. By balancing the new operating condition sample fitting and the parameter update amplitude, the updated model parameters are obtained. The small batch gradient descent algorithm is used to optimize the incremental learning objective function to obtain a battery cell failure prediction model that is adapted to the new operating conditions.

8. A deep learning device for predicting cell failure in a BMS, which is used to implement the deep learning method for predicting cell failure in a BMS according to any one of claims 1 to 7, characterized in that: include: Data acquisition module, feature extraction module, GRU model module, function expansion module and incremental learning module; The data acquisition module is used to collect multi-modal battery cell operation data, pre-process the collected multi-modal battery cell operation data, and divide it into time series samples in a sliding window manner; The feature extraction module is used to extract the time domain features, frequency domain features and differential features of the time series samples, perform feature selection using the Pearson correlation coefficient and mutual information, and perform dimensionality reduction processing using the principal component analysis method to construct a data set; The GRU model module is used to build a cell failure prediction model based on the gated recurrent unit (GRU) model, train the GRU model through data set data, add an attention mechanism to the GRU model, train the GRU model, and optimize model parameters; The function extension module is used to expand the battery cell failure prediction model, predict the battery health status and failure time while predicting the battery cell failure, and introduce an adaptive failure threshold mechanism; The incremental learning module constructs an incremental learning framework based on sample selection for new operating conditions generated when the battery cell is working, and adjusts the battery cell failure prediction model through new operating condition samples.

9. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores a computer program that can be called by the processor; the processor executes the deep learning method for predicting battery cell failure in the BMS according to any one of claims 1 to 7 by calling the computer program stored in the memory.

10. A computer-readable storage medium, characterized in that: Instructions are stored, and when the instructions are executed on a computer, the computer executes the deep learning method for predicting battery cell failure in a BMS as described in any one of claims 1 to 7.

Images