A battery life early prediction method based on pre-classification and LSTM-CNN

Abstract

A battery life early prediction method based on pre-classification and LSTM-CNN, which uses capacity difference data during discharging process to extract features, determines the classification interval of samples through unsupervised learning clustering method, divides into long life group and short life group according to the actual life distribution of the battery, constructs and trains a new superposition model of LSTM and CNN network suitable for battery sample classification and life prediction, first classifies the new sample through the LSTM-CNN classification model, matches different life prediction models, and then predicts the battery life through the LSTM-CNN prediction model, realizes the early prediction of battery life by using a small amount of data, the model structure is simplified, and the prediction accuracy and efficiency are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of battery life prediction technology, and in particular to an early battery life prediction method based on pre-classification and LSTM-CNN. Background Technology

[0002] Batteries, as energy storage devices, have wide and important applications in new energy vehicles, smart grids, and portable electronic devices. However, during their lifespan, batteries experience a gradual decline in electrical performance due to material aging, interface defects, and irreversible internal reactions. Typical performance degradation manifests as capacity decay and power reduction, i.e., a reduction in battery life. Accurately predicting remaining battery life not only helps users rationally plan equipment replacement cycles and reduce operating costs, but also provides early warnings of potential safety risks, improving the overall stability of the system. Traditional battery life prediction methods largely rely on theoretical models and empirical formulas, which typically make very strict assumptions and are difficult to accurately reflect battery degradation mechanisms under actual operating conditions.

[0003] In recent years, with the development of big data and artificial intelligence technologies, more and more research has turned to using real-time monitoring data during battery operation for life prediction. Nevertheless, data-driven methods for battery life prediction still face many challenges. For example, while LSTM models excel at capturing long-term dependencies in time-series data, they are not particularly effective at extracting local features. This is because LSTMs are designed to maintain long-term memory of previous information using their internal memory units and gating mechanisms, making them better at identifying and predicting important events or patterns with long time intervals in sequential data. Autoregressive models (AR models), which use linear combinations of time-series data for prediction, have been applied to estimate remaining battery life (RUL). Their advantages include model simplicity, low computational cost, and small sample requirements, but they also suffer from limited generalization ability and difficulty in capturing nonlinearities and uncertainties in the battery aging process. Convolutional neural networks (CNNs) are mainly used in battery aging modeling because they can automatically extract useful features from high-dimensional battery data. Although they have limitations in capturing the temporal features of battery aging, combining them with models such as LSTM can effectively improve the accuracy of RUL prediction. Currently, models combining Convolutional Neural Networks (CNNs) and Long Short-Term Memory Networks (LSTMs) have shown good accuracy in battery life prediction. However, there is still significant room for optimization in model structure and feature extraction methods to reduce model complexity and improve prediction efficiency and practicality. Patent CN116879753B discloses a battery life prediction method based on big data, proposing a multi-dimensional feature-based battery life prediction method combining Bi-LSTM and XGboost models. This method overcomes the shortcomings of single models and improves the accuracy and robustness of prediction results. However, due to the high time and space complexity of the XGboost model, the overall computational requirements of the prediction model are high, making its implementation difficult. The paper "Tang Y, Yang K, Zheng H, et al. Early prediction of lithium-ion battery lifetime via a hybrid deep learning model[J]. Measurement, 2022, 199: 111530." combines LSTM and CNN networks, proposing a PCLN model, including a data processing module, an LSTM extractor, a CNN extractor, and a prediction layer. Using early-stage battery degradation discharge data as model input yielded good prediction results, but the prediction results for long-life samples were generally low when the sample lifetimes were relatively dispersed. Therefore, reducing the complexity of the fusion model, improving the accuracy of early predictions, and achieving personalized predictions for individual samples are of great significance. Summary of the Invention

[0004] To overcome the limitations of traditional battery life prediction methods in accurately predicting early battery life, which require large amounts of data, have insufficient prediction accuracy, and involve complex model structures, this invention proposes a battery life prediction method based on pre-classification and LSTM-CNN. This method extracts features from capacity difference data during the discharge process, determines the classification intervals of samples through unsupervised learning clustering, and divides the batteries into long-life and short-life groups based on their actual life distribution. A novel superimposed model of LSTM and CNN networks, suitable for battery sample classification and life prediction, is constructed and trained. First, the LSTM-CNN classification model classifies new samples and matches them with different life prediction models. Then, the LSTM-CNN prediction model predicts the battery life. This method achieves early prediction of battery life using limited data, has a simplified model structure, and significantly improves prediction accuracy and efficiency.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for early battery lifetime prediction based on pre-classification and LSTM-CNN includes the following steps:

[0007] S1: First, extract the QV data for each discharge cycle from the collected battery charge and discharge cycle data, and then perform preprocessing on the extracted data.

[0008] S2: Based on data preprocessing, feature extraction and feature selection are performed. The selected features will be used as input for subsequent battery grade classification and life prediction models.

[0009] S3: Using the Mean Shift unsupervised learning clustering method, the battery dataset is divided into two distinct lifespan intervals based on the actual distribution of battery lifespan: a long lifespan group and a short lifespan group. The clustering results will serve as the basis for sample grouping to guide subsequent battery classification and lifespan prediction.

[0010] S4: Build an LSTM-CNN classification model framework, using the features extracted from the training set data as input data, and the grouping situation shown by the clustering results in step S3 as label data, to train the battery life classification model.

[0011] S5: Build an LSTM-CNN regression model framework, and train the corresponding battery life prediction model using the feature data of the long-life group and short-life group batteries in the training set according to the clustering results described in step S3.

[0012] S6: Use test set data to validate and evaluate the model's prediction performance.

[0013] In step S1, during the data preprocessing operation, redundant invalid data outside the discharge process is first identified and removed, and only valid data reflecting the changes in the battery's intrinsic characteristics during the discharge process are retained. On this basis, the first-order difference method is used to detect and identify possible abnormal discharge cycles and abnormal data points in individual cycles. Then, the linear interpolation method is used to reasonably repair these abnormal values. Finally, the voltage data is standardized to unify the voltage sampling points and sampling intervals, so that the data between different batteries are comparable.

[0014] In step S2, during feature extraction and screening, for each charge-discharge cycle, the Q-Q1 data in each cycle is first calculated, and seven basic feature parameters are extracted, including variance, minimum, maximum, average, kurtosis, skewness, and the capacity difference corresponding to a voltage of 2V. Then, by calculating the three correlation coefficients between the seven feature parameters and the total battery capacity (Pearson correlation coefficient, Spearman rank correlation coefficient, and Kendall rank correlation coefficient), the correlation between the seven feature parameters and battery capacity decay is quantitatively analyzed. Subsequently, four feature indicators closely related to battery life are selected for input to subsequent model training and prediction. The selection principle is to select feature items with correlation coefficients greater than 0.9.

[0015] In the lifespan interval division process of step S3, the Mean shift clustering algorithm combined with kernel density estimation is used to perform unsupervised learning clustering on the original battery lifespan data. The specific method is as follows: First, the probability density of the sample lifespan distribution in the dataset is estimated using the Gaussian kernel function. Then, the Mean shift algorithm is used to calculate the drift vector of the sample points. Each sample point is moved in the direction of increasing probability density. The calculation of probability density and the movement of sample points are repeated until the maximum number of iterations is reached. Finally, the battery samples will converge to two points with the largest local density, i.e., cluster centers. Based on the clustering of the long lifespan group and the short lifespan group, the classification intervals, i.e., the long lifespan group and the short lifespan group, are obtained as the label basis for the subsequent classification model.

[0016] Step S4 specifically involves:

[0017] A composite deep learning classification model structure integrating Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) is constructed, including two LSTM layers, one Dropout layer, five one-dimensional convolutional layers, one max pooling layer, one flattening layer, one fully connected layer, and one output layer. The output layer is assigned the Softmax activation function, which transforms the output into a probability distribution. During model training, the clustering results from step S3 are first used to label the training data. The feature data of the first 10%-15% of the discharge data in the training set are used as input to train a classification model, namely the LSTM-CNN classification model. This model classifies the battery into the corresponding lifespan interval based on its current health status, so that different lifespan prediction models can be used to predict the remaining lifespan later.

[0018] Step S5 specifically involves:

[0019] First, an LSTM-CNN regression model framework with the same hierarchical structure as the LSTM-CNN classification model described in step S4 is constructed. It consists of two LSTM layers, five one-dimensional convolutional layers, one fully connected layer, and one output layer. The output layer uses the Linear Activation Function, and the output result is the lifetime value predicted by the model. During the training of the LSTM-CNN regression model, based on the clustering results, the model is trained separately for long-life group and short-life group samples. The first 10%-15% of the loop feature data of the long-life group and short-life group samples are input separately, and different model parameters are assigned to them to capture the battery degradation pattern in different lifetime intervals and to perform accurate lifetime prediction for different batteries. After training, the long-life LSTM-CNN prediction model and the short-life LSTM-CNN prediction model are obtained.

[0020] Step S6 specifically involves:

[0021] First, the feature data of battery samples in the test set is input into the LSTM-CNN classification model for pre-classification to obtain the classification results of the battery samples. Then, based on the pre-classification results, the corresponding LSTM-CNN life prediction model is called for the long-life group / short-life group samples, and the feature data is input to predict the remaining battery life. The root mean square error and mean absolute percentage error are used as evaluation indicators to compare the remaining life predicted by the model with the actual battery life.

[0022] The present invention also includes a battery life early prediction system based on pre-classification and LSTM-CNN, comprising a processing instruction module for performing each step of the aforementioned battery life early prediction method based on pre-classification and LSTM-CNN.

[0023] A device for early battery life prediction based on pre-classification and LSTM-CNN includes:

[0024] Memory: A computer-readable device that stores the computer program for the above-mentioned method of early prediction of battery life based on pre-classification and LSTM-CNN;

[0025] Processor: Used to implement the aforementioned method for early battery life prediction based on pre-classification and LSTM-CNN when executing the computer program.

[0026] A computer-readable storage medium storing a computer program that, when executed by a processor, enables the implementation of the aforementioned method for early prediction of battery life based on pre-classification and LSTM-CNN.

[0027] Compared with the prior art, the advantages of the present invention are:

[0028] (1) Since step S2 adopts the feature extraction method based on the loss mechanism of lithium-ion battery negative electrode active material, calculates Q-Q1 and uses the statistical feature extraction and feature screening method based on correlation coefficient to extract multidimensional strongly correlated features based on the calculation of capacity difference in the discharge process as model input, identifies early aging features of battery, optimizes feature space, reduces dimensional disaster risk, reduces data requirements to achieve early life prediction while improving the prediction accuracy of data-driven model.

[0029] (2) Since step S3 adopts the unsupervised learning method of Mean shift clustering, the classification boundary is defined according to the distribution of the dataset itself, which provides more flexible and reasonable classification rules for sample pre-classification in the subsequent step S4. It has the advantage that the classification boundary matches the actual engineering situation and can improve the accuracy of the prediction results.

[0030] (3) Since steps S4 and S5 adopt the method of constructing an LSTM-CNN fusion model framework, the time series features and complex patterns of battery cycle data can be effectively captured. The learning ability of LSTM units to long time series dependencies is combined with the efficiency of CNN in feature extraction, which significantly improves the accuracy and robustness of classification and prediction models.

[0031] (4) Since step S6 adopts a personalized prediction method that combines pre-classification, it matches the corresponding long-life LSTM-CNN prediction model or short-life LSTM-CNN prediction model for long-life and short-life samples, which has the advantages of refined, personalized and high-precision battery life prediction.

[0032] In summary, this invention employs a feature extraction method based on the loss mechanism of lithium-ion battery anode active materials, combined with unsupervised learning clustering and pre-classification strategies, to construct an LSTM-CNN classification model and an LSTM-CNN prediction model, enabling early personalized prediction of battery life. It has the advantages of simplified model structure, small data requirements, strong generalization performance, and high prediction accuracy. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the overall framework of the lifetime prediction method of the present invention.

[0034] Figure 2 This is a diagram showing the results of Mean-shift clustering.

[0035] Figure 3 This is a schematic diagram of the LSTM-CNN fusion model.

[0036] Figure 4 This is the confusion matrix for the classification model.

[0037] Figure 5 This is a scatter plot of the predicted results and the actual test values. Detailed Implementation

[0038] The present invention will now be described in detail with reference to the accompanying drawings.

[0039] This invention proposes a battery life prediction method based on pre-classification and deep learning. It constructs novel LSTM-CNN classification and regression models suitable for battery life prediction, and combines data preprocessing, feature extraction and filtering, and Mean Shift clustering to accurately predict battery life. Figure 1 As shown, this method enables early, accurate, efficient, and reliable predictions even with limited data. The specific implementation details are as follows.

[0040] A method for early battery lifetime prediction based on pre-classification and LSTM-CNN includes the following steps:

[0041] S1: First, extract the QV data for each discharge cycle from the collected battery charge-discharge cycle data, and then perform preprocessing on the extracted data; the specific steps are as follows:

[0042] S11: Collect a set of capacity-voltage (QV) data for the battery during all charge-discharge cycles throughout its full life cycle, ensuring that the data covers enough information to fully characterize the battery performance;

[0043] The complete life cycle refers to all charge-discharge cycles before the battery capacity degrades to 80% of its rated capacity;

[0044] S12: Clean the collected data and remove redundant data from the non-discharge stage, such as data from the charging stage and other atypical working conditions before and after the discharge process, to ensure that only valid data reflecting the battery discharge characteristics are retained.

[0045] S13: Standardize the voltage data, unify the voltage sampling points and sampling intervals, and ensure that the data of different batteries are comparable under the same sampling conditions;

[0046] S14: Apply the first-order difference method to analyze the effective (QV) data to identify and identify possible abnormal discharge cycles (e.g., abnormally rapid capacity decline or recovery) and abnormal data points within a single cycle.

[0047] Furthermore, the first-order difference calculation method is ΔX i =X i+1 -X i The trend component in the data can be eliminated by first-order difference. If outlier data appears in the calculation results, it means that the original sequence has undergone mutation or abnormality. Based on this, the cycle with abnormal data is marked.

[0048] S14: For any abnormal data points discovered, perform linear interpolation using capacity data with the same voltage from previous and subsequent periods to reasonably repair the data, ensuring the consistency and continuity of the data sequence.

[0049] S15: Finally, standardize the voltage data, unify the voltage sampling points and sampling intervals, so that the data between different batteries are comparable.

[0050] S2: Based on data preprocessing, feature extraction and feature selection are performed. The selected features will be used as input for subsequent battery rating classification and lifespan prediction models. The specific steps are as follows:

[0051] S21: For each battery cycle, calculate the Q-Q1 data (i.e., the capacity difference between each sampling point in the current cycle and the first cycle) to reflect the changes in battery performance between cycles. First, calculate seven basic characteristic parameters for each cycle: variance, minimum, maximum, average, kurtosis, skewness, and the capacity difference at 2V.

[0052] S22: Calculate the three correlation coefficients between the above seven characteristic parameters and the total battery capacity. The three correlation coefficients of the total battery capacity are: Pearson correlation coefficient, Spearman rank correlation coefficient, and Kendall rank correlation coefficient, in order to quantify the degree of correlation between the characteristic parameters and the battery capacity decay.

[0053] Furthermore, the Pearson correlation coefficient is used to measure the degree of linear correlation between the feature parameter and the capacity, while the Spearman rank correlation coefficient and Kendall rank correlation coefficient are used to measure the rank correlation between the feature parameter and the capacity, without considering the actual numerical value of the variable. Their calculation methods are as follows:

[0054]

[0055]

[0056]

[0057] Where n is the sample size, d is the rank difference of the variable, and C and D are the number of consistent pairs and inconsistent pairs, respectively;

[0058] S23: Quantitatively analyze the correlation between seven characteristic parameters and battery capacity decay, and then select four characteristic indicators closely related to battery life for subsequent model training and prediction input. The selection principle is: select characteristic items with correlation coefficients greater than 0.9.

[0059] S3: To refine the battery life prediction problem, the Mean Shift unsupervised learning clustering method is used. Based on the actual distribution of battery life in the dataset, the batteries are divided into two distinct lifespan intervals: a long-life group and a short-life group. The clustering results will serve as the basis for sample grouping to guide subsequent battery classification and lifespan prediction. The specific method is as follows:

[0060] S31: To refine the battery life prediction problem, kernel density estimation is performed on the battery life data. A Mean Shift clustering algorithm combined with kernel density estimation is used for unsupervised learning clustering of the original battery life data. First, a Gaussian kernel function is selected. By transforming the local density around each data point into a probability density function, the overall shape of the data distribution is depicted. The calculation methods for the kernel function and the probability density estimate for each data point are as follows:

[0061]

[0062]

[0063] Where n is the total number of data points, h is the bandwidth, and d is the data dimension;

[0064] S32: Clustering is performed using the Mean Shift clustering algorithm based on the kernel density estimation map. First, each data point is initialized as a candidate cluster center. For each data point, its gradient ascent direction under the kernel density function is calculated, i.e., it moves towards directions with higher density. After each data point moves, its position is updated. This process is repeated until the data point positions converge. Each convergence point represents the center of a cluster. Figure 2 As shown, all batteries are automatically divided into two distinct lifespan ranges: a long lifespan group and a short lifespan group.

[0065] S33: Combining the clustering results, assign a category label to each battery sample, indicating whether it belongs to the long-lifetime or short-lifetime range, as the labeling basis for subsequent classification models. This pre-classification strategy can effectively guide the subsequent model training and prediction process.

[0066] S4: Build an LSTM-CNN classification model framework, using features extracted from the training set data as input data, and the grouping results shown in step S3 as label data, to train the battery life classification model; the specific method is as follows:

[0067] S41: Construct a composite deep learning classification model structure using the Keras framework that integrates a Long Short-Term Memory (LSTM) network and a Convolutional Neural Network (CNN), such as... Figure 3 As shown, it includes:

[0068] Two LSTM layers are used to capture long-range dependencies in the training data, improving the model's ability to handle temporal features. The data input dimension is (number of cycles, number of features), and the number of hidden units is 64.

[0069] A Dropout layer: Drops out a certain percentage of neurons to prevent overfitting; in this example, the dropout percentage is 0.3.

[0070] Five one-dimensional convolutional layers (Conv1D): The first layer has 32 kernels and a kernel width of 16, using the ReLU activation function; the next four layers have 64 kernels and a kernel width of 4, also using the ReLU activation function.

[0071] A max pooling layer: The pooling window size is 8, which effectively reduces data complexity and preserves local features;

[0072] A flatten layer: flattens multidimensional data into one-dimensional data, making it easier for fully connected layers to process it;

[0073] A fully connected layer: using 100,000 neurons to extract more complex and abstract features;

[0074] One output layer: using the Softmax activation function, the model outputs a probability distribution for each category (long lifespan or short lifespan), thus classifying the battery lifespan range.

[0075] S42: Use the preprocessed and feature-extracted data as the model input, the clustering results as the label data, and perform one-hot encoding on the labels. Then, randomly divide the data into training set and test set, and set the proportion of test set to 20%.

[0076] S43: Perform model compilation and training. Set the loss function to cross-entropy loss, use the Adam optimizer to automatically adjust the learning rate, update the model weights, set the model training batch size to 16, the model iteration count to 500, and train the model using the input training set data;

[0077] S44: Input the test set data into the trained model to obtain the predicted classification result, such as... Figure 4 As shown, the prediction accuracy rate is 88.89%. Finally, the prediction model is output and saved.

[0078] S5: Build an LSTM-CNN regression model framework. Based on the clustering results described in step S3, train the corresponding battery life prediction models using the feature data of the long-life group and the short-life group batteries in the training set. The specific method is as follows:

[0079] S51: Similarly, construct two LSTM-CNN regression models with similar structures to the classification model described in step S41. The structures of each layer before the output layer are the same, but the activation function of the output layer is changed to a linear activation function to directly output the remaining lifetime prediction value. The two models are used for lifetime prediction of long-lived samples and short-lived samples, respectively.

[0080] S52: Perform model compilation and training. Set the loss function to root mean square error loss, use the Adam optimizer to automatically adjust the learning rate, set the model training batch size to 4, and the model iteration count to 1000. Divide the training set data into long-life and short-life portions according to the lifespan intervals shown in the clustering results. Train two independent lifespan prediction models using the feature data of long-life and short-life batteries respectively. Optimize the model parameters according to the characteristics of each sample to more accurately capture the battery degradation patterns in different lifespan intervals.

[0081] S6: For new samples, follow steps S1-S5 to predict battery life. Use test set data to validate and evaluate the model's prediction performance, i.e., validate and evaluate the overall model's prediction effectiveness. The specific method is as follows:

[0082] S61: First, call the trained classification model described in step S4, take the feature data of the test set samples as the model input, obtain its pre-classification result (i.e., the prediction of the lifespan interval), and divide the test set samples into long lifespan group and short lifespan group.

[0083] S62: Based on the classification results, call the lifespan prediction model described in step S5, and again use the feature data as input to predict the remaining lifespan of the test sample using both the long-lifespan prediction model and the short-lifespan prediction model, and output the predicted value of the sample battery life, such as... Figure 5 As shown;

[0084] S63: To quantify the model's predictive performance, root mean square error and mean absolute percentage error are used as evaluation metrics. The model's predicted remaining lifespan is compared with the actual battery lifespan to verify the model's predictive performance and accuracy.

[0085] The calculation method for the evaluation indicators is as follows:

[0086]

[0087]

[0088] Based on the prediction results and actual battery life data, the root mean square error (RMSE) was calculated to be 150.94, and the mean percentage error (MPI) was 12.5%. The prediction performance was particularly good for the short-life group, with an RMS of 75.79 and an MPI of 10.61%, indicating that this method can effectively utilize limited data to achieve early prediction of battery life and achieve high prediction accuracy, especially for short-life battery samples, where the method provides more accurate prediction results. In summary, this method constructs a simpler LSTM-CNN composite model, combined with a pre-classification strategy, enabling early personalized predictions for both long-life and short-life batteries with limited data, thus improving prediction accuracy and robustness.

[0089] The present invention also includes a battery life early prediction system based on pre-classification and LSTM-CNN, comprising a processing instruction module for performing each step of the aforementioned battery life early prediction method based on pre-classification and LSTM-CNN.

[0090] A device for early battery life prediction based on pre-classification and LSTM-CNN includes:

[0091] Memory: A computer-readable device that stores the computer program for the above-mentioned method of early prediction of battery life based on pre-classification and LSTM-CNN;

[0092] Processor: Used to implement the aforementioned method for early battery life prediction based on pre-classification and LSTM-CNN when executing the computer program.

[0093] A computer-readable storage medium storing a computer program that, when executed by a processor, enables the implementation of the aforementioned method for early prediction of battery life based on pre-classification and LSTM-CNN.

[0094] In summary, this invention addresses the problems of low early prediction accuracy, complex model structure, and large data usage in traditional battery life prediction methods. It proposes a new method for early battery life prediction based on pre-classification and LSTM-CNN (Long Short-Term Memory-Convolutional Neural Network) deep learning technology. First, effective QV data reflecting the battery's intrinsic characteristics is extracted from battery charge-discharge cycle data. Second, seven features highly correlated with battery life are calculated, including variance, minimum, maximum, average, kurtosis, skewness, and the capacity difference corresponding to a voltage of 2V, for feature extraction, and the data are filtered by calculating correlation coefficients. Based on this, a Mean Shift clustering algorithm combined with kernel density estimation is used to perform unsupervised learning clustering of the battery life data, dividing the battery samples into two significantly discriminative lifespan intervals: a long-life group and a short-life group. Then, a composite deep learning model integrating LSTM and CNN is designed and developed for battery sample classification and specific battery life prediction, respectively. The LSTM-CNN classification model is used to identify battery health status and classify them into corresponding lifespan intervals. Based on the pre-classification results, the LSTM-CNN regression model uses independently trained lifespan prediction models to make final lifespan predictions for long-life and short-life batteries respectively. Finally, the root mean square error (RMSE) and mean absolute percentage error (MAPE) are used to evaluate the predictive performance of the models. This invention achieves accurate early prediction of battery lifespan using only a small amount of finite-cycle discharge data. Furthermore, the independently trained lifespan prediction models provide more refined predictions for long-life and short-life batteries. In addition, the model structure is relatively simple, easy to implement and maintain, and significantly improves the efficiency, accuracy, reliability, and economy of battery prediction.

Claims

1. A method for early prediction of battery life based on pre-classification and LSTM-CNN, characterized in that, Includes the following steps: S1: First, extract the QV data for each discharge cycle from the collected battery charge and discharge cycle data, and then perform preprocessing on the extracted data. S2: Based on data preprocessing, feature extraction and feature selection are performed. The selected features will be used as input for subsequent battery grade classification and life prediction models. S3: Using the Mean Shift unsupervised learning clustering method, the battery dataset is divided into two distinct lifespan intervals based on the actual distribution of battery lifespan: a long lifespan group and a short lifespan group. The clustering results will serve as the basis for sample grouping to guide subsequent battery classification and lifespan prediction. S4: Build an LSTM-CNN classification model framework, using the features extracted from the training set data as input data, and the grouping situation shown by the clustering results in step S3 as label data, to train the battery life classification model. S5: Build an LSTM-CNN regression model framework, and train the corresponding battery life prediction model using the feature data of the long-life group and short-life group batteries in the training set according to the clustering results described in step S3. S6: Use test set data to validate and evaluate the model's prediction performance; Step S4 specifically involves: A composite deep learning classification model structure integrating Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) is constructed, including two LSTM layers, one Dropout layer, five one-dimensional convolutional layers, one max pooling layer, one flattening layer, one fully connected layer, and one output layer. The output layer specifies the Softmax function as the activation function, transforming the output into a probability distribution. During model training, the training data is first labeled using the clustering results from step S3. The feature data of the first 10%-15% of the cyclic discharge data in the training set are used as input to train a classification model, namely the LSTM-CNN classification model. This model classifies the battery into the corresponding lifespan interval based on its current health condition, so that different lifespan prediction models can be used to predict the remaining lifespan in the future. Step S5 specifically involves: First, an LSTM-CNN regression model framework with the same hierarchical structure as the LSTM-CNN classification model described in step S4 is constructed. It consists of two LSTM layers, five one-dimensional convolutional layers, one fully connected layer, and one output layer. The output layer uses the Linear Activation Function, and the output result is the lifetime value predicted by the model. During the training of the LSTM-CNN regression model, based on the clustering results, the model is trained separately for long-life group and short-life group samples. The first 10%-15% of the iterative feature data of the long-life group and short-life group samples are input separately, and different model parameters are assigned to them to capture the battery degradation pattern in different lifetime intervals and to perform accurate lifetime prediction for different batteries. After training, the long-life LSTM-CNN prediction model and the short-life LSTM-CNN prediction model are obtained.

2. The method for early battery lifetime prediction based on pre-classification and LSTM-CNN according to claim 1, characterized in that, In step S1, during the data preprocessing operation, redundant invalid data outside the discharge process is first identified and removed, and only valid data reflecting the changes in the battery's intrinsic characteristics during the discharge process are retained. On this basis, the first-order difference method is used to detect and identify the existing abnormal discharge cycles and abnormal data points in individual cycles. Then, the linear interpolation method is used to reasonably repair these abnormal values. Finally, the voltage data is standardized to unify the voltage sampling points and sampling intervals, so that the data between different batteries are comparable.

3. The method for early battery lifetime prediction based on pre-classification and LSTM-CNN according to claim 1, characterized in that, In step S2, during feature extraction and screening, for each charge-discharge cycle, the Q-Q1 data in each cycle is first calculated, and seven basic feature parameters are extracted, including variance, minimum, maximum, average, kurtosis, skewness, and the capacity difference corresponding to a voltage of 2V. Then, by calculating the three correlation coefficients between the seven feature parameters and the total battery capacity (Pearson correlation coefficient, Spearman rank correlation coefficient, and Kendall rank correlation coefficient), the correlation between the seven feature parameters and battery capacity decay is quantitatively analyzed. Subsequently, four feature indicators closely related to battery life are selected for input to subsequent model training and prediction. The selection principle is to select feature items with correlation coefficients greater than 0.

9.

4. The method for early battery lifetime prediction based on pre-classification and LSTM-CNN according to claim 1, characterized in that, In the lifespan interval division process of step S3, the Mean shift clustering algorithm combined with kernel density estimation is used to perform unsupervised learning clustering on the original battery lifespan data. The specific method is as follows: First, the probability density of the sample lifespan distribution in the dataset is estimated using the Gaussian kernel function. Then, the Mean shift algorithm is used to calculate the drift vector of the sample points. Each sample point is moved in the direction of increasing probability density. The calculation of probability density and the movement of sample points are repeated until the maximum number of iterations is reached. Finally, the battery samples will converge to two points with the largest local density, i.e., cluster centers. Based on the clustering of the long lifespan group and the short lifespan group, the classification intervals, i.e., the long lifespan group and the short lifespan group, are obtained as the label basis for the subsequent classification model.

5. The method for early prediction of battery lifetime based on pre-classification and LSTM-CNN according to claim 1, characterized in that, Step S6 specifically involves: First, the feature data of battery samples in the test set is input into the LSTM-CNN classification model for pre-classification to obtain the classification results of the battery samples. Then, based on the pre-classification results, the corresponding LSTM-CNN life prediction model is called for the long-life group / short-life group samples, and the feature data is input to predict the remaining battery life. The root mean square error and mean absolute percentage error are used as evaluation indicators to compare the remaining life predicted by the model with the actual battery life.

6. A battery life early prediction system based on pre-classification and LSTM-CNN, characterized in that, It includes a processing instruction module for performing each step of the battery life early prediction method based on pre-classification and LSTM-CNN as described in any one of claims 1-5.

7. A device for early prediction of battery life based on pre-classification and LSTM-CNN, characterized in that, include: Memory: A computer-readable device storing the computer program of the battery life early prediction method based on pre-classification and LSTM-CNN as described in any one of claims 1-5; Processor: Used to implement the battery life early prediction method based on pre-classification and LSTM-CNN as described in any one of claims 1-5 when executing the computer program.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of a battery life early prediction method based on pre-classification and LSTM-CNN as described in any one of claims 1-5.

Images