Lithium battery performance prediction method and device based on deep learning, equipment and medium

Abstract

The invention is suitable for the technical field of lithium batteries, and provides a lithium battery performance prediction method based on deep learning, and the method comprises the steps: carrying out the fine tuning training of a trained lithium battery performance prediction model through the early cycle data of a target battery, performing online prediction on the performance of the target battery by using the lithium battery performance prediction model subjected to fine tuning training to obtain the residual service life and the health state of the target battery, and screening a preset battery sample data set by adopting a sample screening strategy before fine tuning training is performed on the lithium battery performance prediction model to obtain the residual service life and the health state of the target battery. According to the method, the training sample set is obtained, and the training sample set is utilized to perform offline training on the pre-constructed lithium battery performance prediction model to obtain the trained lithium battery performance prediction model, so that the prediction precision and the prediction effect of lithium battery performance prediction are improved, and accurate prediction of the lithium battery capacity degradation track is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium batteries, and in particular relates to a lithium battery performance prediction method, device, equipment and medium based on deep learning. Background Art

[0002] Lithium batteries are widely used in various 3C products and are one of the most widely used mobile energy storage systems. Lithium batteries release and store energy by embedding and de-embedding lithium ions between the positive and negative electrodes of the battery. However, this chemical process is not completely reversible. As the number of battery cycles increases, the microstructure of the positive and negative electrodes will undergo irreversible changes, causing permanent degradation of the macroscopic capacity of the lithium battery until the battery failure threshold is reached. Accidental battery failure may not only cause permanent damage to electrical appliances, but may also cause safety problems. Therefore, it is very necessary to predict the remaining useful life (RUL) and state of health (SOH) of lithium batteries in advance.

[0003] At present, the main prediction methods for lithium batteries are divided into physical and chemical model methods and data-driven methods. For physical and chemical model methods, most of the work is based on three traditional methods: electrochemical model, equivalent circuit model, and empirical model of the battery. These methods start from the physical and chemical properties of the battery, capture model parameters according to the cycle performance of the battery, and then make predictions, so they have good interpretability. However, these models themselves are very complex, and the determination of experimental parameters often requires a large number of experiments, and the accuracy of the model is low. For data-driven methods, since they often only focus on the real cycle data of the battery and do not require special additional experiments to determine specific parameters, they save time and efficiency, and have high prediction accuracy, so they are slowly becoming a hot topic. Common data-driven methods mainly consist of three steps: feature production, feature sampling, and time series modeling. Severson et al. extracted the data of the tenth and hundredth cycles of the battery cycle into features, and proved that these features have a strong correlation with the remaining service life of the battery through the Pearson correlation coefficient. Wang et al. used a multivariate long short-term memory network (Multivariate Long Short-Term Memory, Multivariate LSTM) to capture the degradation information of the battery, achieving better prediction results than networks such as LSTM. However, these works strongly rely on the knowledge of the battery field, and the effect of feature creation will directly affect the prediction results. In addition, the machine learning methods used by Severson et al. can often only give a point estimate of the remaining service life end-to-end, but cannot predict the battery capacity degradation trajectory and cannot accurately grasp the battery degradation information. Summary of the invention

[0004] The purpose of the present invention is to provide a lithium battery performance prediction method, device, equipment and storage medium based on deep learning, aiming to solve the problem that the existing technology cannot predict the lithium battery capacity degradation trajectory, resulting in poor lithium battery performance prediction effect.

[0005] In one aspect, the present invention provides a lithium battery performance prediction method based on deep learning, the method comprising the following steps:

[0006] Use the early cycle data of the target battery to fine-tune the trained lithium battery performance prediction model;

[0007] Using the fine-tuned and trained lithium battery performance prediction model to perform online prediction on the performance of the target battery, and obtain the remaining service life and health status of the target battery;

[0008] The lithium battery performance prediction model is trained in the following way:

[0009] A sample screening strategy is used to screen a preset battery sample data set to obtain a training sample set;

[0010] The pre-built lithium battery performance prediction model is trained offline using the training sample set to obtain a trained lithium battery performance prediction model.

[0011] Preferably, the lithium battery performance prediction model includes a dual-channel feature extraction module, a feature aggregation module and a timing processing and prediction module, wherein the dual-channel feature extraction module includes a first channel module and a second channel module, and the first channel module and the second channel module are both embedded with channel attention and spatial attention modules, the feature aggregation module includes a flattening layer and a splicing layer, and the timing processing and prediction module includes a long short-term memory network, two fully connected layers and a timing attention module.

[0012] Preferably, the first channel module and the second channel module each include two convolutional layers, and each convolutional layer is connected to a maximum pooling layer. At the same time, a channel attention and spatial attention module is embedded between each convolutional layer and the maximum pooling layer.

[0013] Preferably, the step of performing offline training on the pre-built lithium battery performance prediction model using the training sample set comprises:

[0014] According to the training sample set, constructing a quasi-Gram angular field matrix and a degradation matrix of battery samples in the training sample set;

[0015] The lithium battery performance prediction model is trained offline using the quasi-Gram angular field matrix and the degradation matrix.

[0016] Preferably, the step of using a sample screening strategy to screen a preset battery sample data set includes:

[0017] Cleaning the battery sample data set according to a preset capacity drop threshold to obtain a remaining sample data set;

[0018] According to the remaining sample data set, extracting the capacity degradation matrix of all battery samples in the remaining sample data set;

[0019] Performing dimension reduction processing on the capacity degradation matrix by using principal component analysis to obtain a capacity dimension reduction matrix;

[0020] Using an unsupervised learning algorithm to perform clustering processing on the capacity dimension reduction matrix to obtain a clustering result;

[0021] Based on the preset prior features and the clustering results, the remaining sample data sets are screened to obtain the training sample set.

[0022] On the other hand, the present invention provides a lithium battery performance prediction device based on deep learning, the device comprising:

[0023] A model fine-tuning unit, used to fine-tune the trained lithium battery performance prediction model using early cycle data of the target battery;

[0024] A performance prediction unit, used to use the fine-tuned and trained lithium battery performance prediction model to perform online prediction on the performance of the target battery, and obtain the remaining service life and health status of the target battery;

[0025] The lithium battery performance prediction model is trained in the following way:

[0026] A sample screening strategy is used to screen a preset battery sample data set to obtain a training sample set;

[0027] The pre-built lithium battery performance prediction model is trained offline using the training sample set to obtain a trained lithium battery performance prediction model.

[0028] Preferably, the lithium battery performance prediction model includes a dual-channel feature extraction module, a feature aggregation module and a timing processing and prediction module, wherein the dual-channel feature extraction module includes a first channel module and a second channel module, and the first channel module and the second channel module are both embedded with channel attention and spatial attention modules, the feature aggregation module includes a flattening layer and a splicing layer, and the timing processing and prediction module includes a long short-term memory network, two fully connected layers and a timing attention module.

[0029] Preferably, the sample screening unit comprises:

[0030] A data cleaning unit, used for cleaning the battery sample data set according to a preset capacity drop threshold to obtain a remaining sample data set;

[0031] A matrix extraction unit, configured to extract, according to the remaining sample data set, a capacity degradation matrix of all battery samples in the remaining sample data set;

[0032] A matrix dimension reduction unit, used for performing dimension reduction processing on the capacity degradation matrix by using principal component analysis to obtain a capacity dimension reduction matrix;

[0033] A matrix clustering unit, used for performing clustering processing on the capacity dimension reduction matrix using an unsupervised learning algorithm to obtain a clustering result;

[0034] The sample screening subunit is used to screen the remaining sample data set based on preset prior features and the clustering result to obtain the training sample set.

[0035] On the other hand, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps described in the above-mentioned method for predicting lithium battery performance based on deep learning are implemented.

[0036] On the other hand, the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps described in the above-mentioned lithium battery performance prediction method based on deep learning.

[0037] The present invention uses the early cycle data of the target battery to perform fine-tuning training on the trained lithium battery performance prediction model, and uses the fine-tuned lithium battery performance prediction model to perform online prediction on the performance of the target battery to obtain the remaining service life and health status of the target battery. Before fine-tuning the lithium battery performance prediction model, a sample screening strategy is used to screen a preset battery sample data set to obtain a training sample set, and the training sample set is used to perform offline training on the pre-built lithium battery performance prediction model to obtain a trained lithium battery performance prediction model, thereby improving the prediction accuracy and prediction effect of the lithium battery performance prediction, and realizing accurate prediction of the lithium battery capacity degradation trajectory. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a flow chart of an implementation method of a lithium battery performance prediction method based on deep learning provided in Embodiment 1 of the present invention;

[0039] Figure 2is a flow chart of the implementation of the lithium battery performance prediction method based on deep learning provided in the second embodiment of the present invention;

[0040] Figure 3 is a schematic diagram of the structure of a lithium battery performance prediction model provided in Embodiment 2 of the present invention;

[0041] Figure 4 It is a structural schematic diagram of a dual-channel feature extraction module in a lithium battery performance prediction model provided in the second embodiment of the present invention;

[0042] Figure 5 is a schematic diagram of the structure of a lithium battery performance prediction device based on deep learning provided in Embodiment 3 of the present invention;

[0043] Figure 6 It is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0045] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments:

[0046] Embodiment 1:

[0047] Figure 1 The implementation process of the lithium battery performance prediction method based on deep learning provided in the first embodiment of the present invention is shown. For the convenience of explanation, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0048] In step S101, the trained lithium battery performance prediction model is fine-tuned using the early cycle data of the target battery.

[0049] The embodiments of the present invention are applicable to electronic devices, such as personal computers, servers, etc. In the embodiments of the present invention, the early cycle data of the target battery (for example, the data of the battery capacity decay 1% stage or the data of the first 50 cycle stages) are obtained. These data may include charge and discharge curves, capacity decay curves, coulomb efficiency, voltage change data, etc. The early cycle data are processed according to the model structure of the lithium battery performance prediction model and the characteristics of the input data, and the processed target data are input into the pre-trained lithium battery performance prediction model to achieve fine-tuning of the model. In addition, the specific implementation of the model structure of the lithium battery performance prediction model and the characteristics of the input data are described in the following embodiment 2, which will not be repeated here.

[0050] In step S102, the performance of the target battery is predicted online using the fine-tuned and trained lithium battery performance prediction model to obtain the remaining service life and health status of the target battery.

[0051] In the embodiment of the present invention, the fine-tuned and trained lithium battery performance prediction model is used to perform online prediction of the next capacity change of the target battery to obtain the remaining useful life (RUL) and health state (SOH) of the target battery, where RUL refers to the time or number of cycles that the lithium battery can still work normally in the current health state, and SOH is the current state of health of the lithium battery to provide specific power and energy. Specifically, the calculation formula RUL is used α =CC EOL -CC α and RUL and SOH are calculated respectively, where C actual is the current capacity of the target battery predicted by the lithium battery performance prediction model, C nom is the initial rated capacity of the target battery, α is the current cycle number, RUL α is the remaining service life of the current number of cycles a, CC EOL The total number of cycles to reach end of life (EOL) is CC α The number of cycles completed to reach the current number of laps.

[0052] In an embodiment of the present invention, the early cycle data of the target battery is used to fine-tune the trained lithium battery performance prediction model, and the fine-tuned lithium battery performance prediction model is used to perform online prediction of the performance of the target battery to obtain the remaining service life and health status of the target battery, thereby improving the prediction accuracy and prediction effect of the lithium battery performance prediction, and achieving accurate prediction of the lithium battery capacity degradation trajectory.

[0053] Embodiment 2:

[0054] Figure 2 The implementation process of the lithium battery performance prediction method based on deep learning provided in the second embodiment of the present invention is shown. For the convenience of explanation, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0055] In step S201, a sample screening strategy is used to screen a preset battery sample data set to obtain a training sample set.

[0056] The embodiments of the present invention are applicable to electronic devices, such as personal computers, servers, etc. In the embodiments of the present invention, the battery sample data set is a collected original battery cycle data set with detailed records of commercial lithium-ion batteries cycled to failure under fast charging conditions. Since the cycle life changes between different batteries vary greatly, directly using all samples to learn and model battery degradation may cause unnecessary burdens, so it is necessary to select appropriate battery samples through appropriate methods. Based on this, a sample screening strategy is used to screen the battery sample data set, that is, to eliminate or selectively retain certain samples in order to more accurately reflect the overall characteristics of the data set or optimize for specific problems (such as lithium battery capacity degradation), so that the training sample set obtained after screening can improve the training efficiency and generalization ability of the deep learning model.

[0057] In a feasible embodiment, an open source data set is used as a battery sample data set, which contains 124 commercial lithium-ion batteries that are cycled to failure under fast charging conditions. These batteries are manufactured by A123 Systems (specific model APR18650M1A), using lithium phosphate (LFP) / graphite as positive and negative electrode materials, and each battery has a nominal capacity of 1.1 Ah and a nominal voltage of 3.3 V. These batteries are cycled in a horizontal cylindrical fixture of a 48-channel Arbin LBT constant potentiostat, and the test temperature is set to 30°C. In addition, all batteries adopt a one-step fast charging or a two-step fast charging strategy, and the charging strategy of the battery is randomly assigned. For one-step fast charging, the battery is first charged to 80% SOC (State Of Charge) at a constant current, and then charged to a cut-off voltage in a constant current-constant voltage (CC-CV) mode of 1C; for two-step fast charging, the battery is first charged to a certain intermediate capacity Q1 at a constant current, and then charged to 80% SOC at a constant current, and finally charged to a cut-off voltage in a constant current-constant voltage (CC-CV) mode of 1C. The CC-CV mode is charged to the cut-off voltage, and all batteries are discharged at a constant current of 4C, with an upper voltage limit of 3.6V and a lower voltage limit of 2.0V. It is worth noting that the 124 lithium batteries in this data set are divided into three batches for testing. The test time interval between these three batches is as long as half a year, and there is noise caused by power outages, device restarts, etc. during the test process.

[0058] In a feasible embodiment, the battery sample data set is screened through the following steps to obtain a training sample set:

[0059] (1) Cleaning the battery sample data set according to a preset capacity drop threshold to obtain a remaining sample data set;

[0060] In an embodiment of the present invention, there are many abnormal capacity drops in the original battery sample data set. These capacity drops are caused by factors such as power failure of the cycling equipment and restart of the equipment during the cycle test. For short-term equipment abnormalities, it manifests as a small capacity drop; for long-term equipment abnormalities, it manifests as a large capacity drop. Long-term equipment abnormalities may cause serious deviations in the battery's cycling performance and affect the overall prediction accuracy. Based on this, the battery sample data set is cleaned according to a preset capacity drop threshold to delete the severely abnormal battery samples in the data set to obtain the remaining sample data set.

[0061] In a feasible embodiment, the capacity drop threshold is set to 0.035Ah. Battery samples with a capacity drop less than this threshold are considered to be tolerable device noise. Battery samples with a capacity drop greater than this threshold are deleted.

[0062] (2) extracting the capacity degradation matrix of all battery samples in the remaining sample data set according to the remaining sample data set;

[0063] In an embodiment of the present invention, the battery capacity degradation pattern refers to the various characteristics and data exhibited by the gradual decline in performance during long-term use by analyzing the battery's charge and discharge signals, so as to perform data screening and establish features. The capacity degradation of lithium batteries is affected by the coupling of multiple factors, such as the influence of the ambient temperature of the battery cycle, the nonlinear degradation characteristics of the battery, the external comprehensive factors of the battery cycle, manufacturing errors, charge and discharge strategies, and differences in the components of the battery itself, which will affect the cycle life of the battery. Therefore, it is difficult to accurately capture the degradation of lithium batteries in different scenarios by directly using the capacity degradation curve of lithium batteries. Based on this, in order to analyze the degradation pattern of lithium batteries, for each battery sample in the remaining sample data set, a capacity degradation matrix is ​​constructed according to the data on the change of its capacity over time. The capacity degradation matrix reflects the trend of the capacity change of the battery during use. Specifically, the capacity degradation matrix S is expressed as Among them, i represents the number of battery samples, j represents the number of cycles of the battery with the longest life, Qi j It represents the capacity of the i-th battery sample at the j-th cycle. In addition, for battery samples with less than j cycles, the matrix is ​​filled with 0.

[0064] (3) Using principal component analysis to reduce the dimension of the capacity degradation matrix, a capacity dimension reduction matrix is ​​obtained;

[0065] In an embodiment of the present invention, principal component analysis (PCA) is performed on the capacity degradation matrix. By selecting to retain 95% of the degradation information, the capacity degradation matrix S is subjected to dimensionality reduction processing to obtain a capacity dimensionality reduction matrix. The capacity dimensionality reduction matrix retains the main change characteristics in the original data while reducing the dimension of the data.

[0066] In a feasible embodiment, when the capacity degradation matrix is ​​reduced in dimension by using principal component analysis by selecting to retain 95% of the degradation information, specifically, first, the covariance matrix of the capacity degradation matrix S is calculated. The covariance matrix reflects the correlation between the features and is the key to PCA analysis. Then, the covariance matrix is ​​subjected to eigenvalue decomposition to obtain eigenvalues ​​and corresponding eigenvectors. The eigenvalues ​​represent the importance of the corresponding eigenvectors (i.e., the size of the variance). In PCA, a larger eigenvalue means that the corresponding principal component (i.e., the eigenvector) has a larger variance contribution in the data set, and therefore contains more information. The eigenvalues ​​are arranged in order from large to small, i.e., the eigenvalue λ is the largest eigenvalue. 1 >=λ 2 >=…>=λ m , then, starting from the largest eigenvalue, the contribution rate of each eigenvalue is accumulated in turn until the cumulative contribution rate reaches or exceeds the specified threshold (such as 0.95), that is, Among them, m is the dimension of the original data (that is, the total number of eigenvalues), k is the number of eigenvalues ​​considered in the accumulation process, when the cumulative contribution rate reaches or exceeds the specified threshold, the corresponding number of eigenvalues ​​k is the retained feature dimension, finally, the eigenvectors corresponding to the first k eigenvalues ​​are used as the basis vectors of the new space, and the original data is projected into this new low-dimensional space to obtain the capacity dimension reduction matrix R after dimensionality reduction, that is, Wherein, i={1, 2, ..., the maximum number of battery samples}, j={1, 2, ..., k}, and the maximum number of battery samples is the total number of battery samples in the remaining sample data set.

[0067] (4) Use an unsupervised learning algorithm to cluster the capacity dimension reduction matrix to obtain a clustering result;

[0068] In the embodiment of the present invention, an unsupervised graph clustering algorithm is used to perform clustering processing on the capacity dimension reduction matrix, so as to divide the battery samples in the remaining sample data set into multiple clusters to obtain a clustering result.

[0069] In a feasible embodiment, spectral clustering (SC) is used to automatically identify degradation patterns of the capacity reduction matrix to obtain three cluster groups, which correspond to different degradation patterns of long-term, medium-term and short-term cycle life, respectively, and each cluster group contains battery samples with similar capacity degradation characteristics.

[0070] (5) Based on the preset prior features and clustering results, the remaining sample data sets are screened to obtain the training sample set.

[0071] In an embodiment of the present invention, although unsupervised learning spectral clustering can be used to classify batteries with different RULs with zero error, batteries with similar RULs do not necessarily have similar degradation patterns. In order to further accurately identify the degradation patterns of different batteries, a priori features of domain knowledge are introduced to evaluate the accuracy of clustering results to assist in classification, wherein a priori features refer to features related to battery degradation signals that are artificially extracted through prior knowledge, thereby identifying battery degradation patterns. Here, three priori features are used as shown in Table 1 below, and the Pearson correlation coefficient is used to evaluate the correlation between these three features and the battery RUL. The closer the Pearson correlation coefficient is to 1, the more positively correlated the feature is with the RUL. The closer the Pearson correlation coefficient is to -1, the more negatively correlated the feature is with the RUL. The closer the Pearson correlation coefficient is to 0, the less correlated the feature is with the RUL. Here, ΔQ 100-10 (V) represents the series obtained by subtracting the linear interpolation voltage in the 100th cycle discharge voltage-capacity curve from the 10th cycle discharge voltage-capacity curve, ΔV 100-10 (Q = 0.85Ah) represents the value obtained by subtracting the voltage closest to Q = 0.85Ah in the battery's 100th cycle discharge voltage-capacity curve from the voltage closest to Q = 0.85Ah in the battery's 10th cycle discharge voltage-capacity curve. As can be seen from Table 1, the three extracted prior features have a correlation of more than 90% with capacity degradation, indicating that degradation pattern recognition based on such artificially extracted features has a high degree of credibility. It is worth noting that these features are all obtained from the first 100 cycles of the battery cycle, indicating that batteries under different charging strategies have significant differences in the early stages of the cycle, providing strong evidence for the rationality of early predictions of batteries.

[0072] Table 1

[0073] Prior features symbol Pearson correlation coefficient <![CDATA[Standard deviation ΔQ 100-10 (V)]]> std(ΔQ) -0.92473 <![CDATA[Average value|ΔQ 100-10 (V)|]]> mean(ΔQ) -0.90906 <![CDATA[|ΔV 100-10 (Q=0.85Ah)|]]> abs(ΔV) -0.90962

[0074] In an embodiment of the present invention, when the remaining sample data set is screened based on the preset prior features and clustering results, the clustering results of the spectral clustering are further screened using the three prior features in Table 1. For the screened clustering results, the K-Nearest Neighbor (KNN) algorithm is used to identify batteries with similar cycle life as the training sample set of the target battery. Specifically, for each cluster grouping of the clustering results, the battery samples in the cluster grouping that do not meet the thresholds corresponding to these prior features are deleted, that is, the battery samples are removed from the remaining sample data set. For the remaining sample data set after the deletion, the KNN algorithm is used to calculate the Euclidean distance between the target battery and each battery sample in the remaining sample data set in the feature space. According to the distance, a specified number of nearest neighbor samples are found as training samples. These samples have a higher similarity with the target battery in terms of capacity degradation, and these training samples constitute the training sample set.

[0075] The battery sample data set is screened through the above steps (1)-(5), thereby achieving efficient and accurate feature extraction by combining domain knowledge and unsupervised learning, effectively reducing redundant data and noise signals in the training sample set, and thus improving the training effect and training speed of subsequent models.

[0076] In step S202, the pre-built lithium battery performance prediction model is offline trained using the training sample set to obtain a trained lithium battery performance prediction model.

[0077] In an embodiment of the present invention, battery samples in a training sample set are input as image features into a pre-built lithium battery performance prediction model, and the lithium battery performance prediction model is trained offline for battery degradation pattern learning to obtain a trained lithium battery performance prediction model.

[0078] In a possible embodiment, if Figure 3 As shown, the lithium battery performance prediction model (CNN-Attention-LSTM) includes a dual-channel feature extraction module, a feature aggregation module, and a timing processing and prediction module, wherein the dual-channel feature extraction module includes a first channel module and a second channel module, and the first channel module and the second channel module are both embedded with channel attention and spatial attention modules, the feature aggregation module includes a flattening layer and a splicing layer, and the timing processing and prediction module includes a long short-term memory network, two fully connected layers and a timing attention module, thereby improving the prediction accuracy of the model.

[0079] In an embodiment of the present invention, a dual-channel feature extraction module is used to perform feature extraction processing on an input matrix, and to process the extracted features through a channel attention and a spatial attention module (Convolutional Block Attention Module, CBAM) to obtain image features; a feature aggregation module is used to flatten and splice the image features output by the dual-channel feature extraction module to obtain a set of one-dimensional feature vectors, which represent the fusion of dual-channel features, wherein the flattening layer is used to flatten the two-dimensional image features axially, and the splicing layer is used to splice the flattened one-dimensional vectors of the two channels together to meet the input conditions of the timing network; the timing processing and prediction module is used to perform timing modeling representation on the aggregated one-dimensional feature vectors, wherein the timing attention module is also used to improve the modeling effect, and the final result is predicted, wherein the long short-term memory network (Long Short-Term Memory, LSTM) is used to perform timing modeling on the feature vectors obtained by the feature aggregation module to obtain the timing law of each original image feature, the timing attention module is used to improve the timing modeling effect, and the fully connected layer module is used to integrate the module for timing output.

[0080] In another possible embodiment, Figure 4 As shown, the first channel module and the second channel module each include two convolutional layers (Conv) for feature extraction, and each convolutional layer is connected to a maximum pooling layer (MaxPool) to prevent network overfitting. At the same time, a channel attention and spatial attention module (CBAM) is embedded between each convolutional layer and the maximum pooling layer to accelerate training by filtering features through CBAM. The convolution kernel size of the two Conv layers is 3*3, and the pooling window size of the two MaxPool layers is 2*2.

[0081] In another feasible embodiment, based on the training sample set, a quasi-Gram angular field matrix and a degradation matrix of the battery samples in the training sample set are constructed; the quasi-Gram angular field matrix and the degradation matrix are used to perform offline training on the lithium battery performance prediction model, thereby achieving accurate prediction of the capacity degradation trajectory.

[0082] In an embodiment of the present invention, according to the training sample set, the quasi-Gramian Angular Field (GAF) matrix and the degradation matrix (DM) of the battery samples in the training sample set are first constructed to capture the fluctuation characteristics of the battery through GAF and the long-term degradation trend characteristics of the battery through DM. Then, GAF is input into the first channel module of the lithium battery performance prediction model, and DM is input into the second channel module to perform offline training on the lithium battery performance prediction model. Specifically, when constructing the GAF matrix of capacity, first, the degradation matrix C of each battery sample in the training sample set is converted into [C 1 ,C 2 ,…,C n ] is normalized to Then, using the polarization equation right After polar coordinate conversion, according to the equation The GAF matrix is ​​obtained, where each point of the GAF matrix is ​​represented as the nonlinear relationship between the i-th cycle capacity and the j-th cycle capacity of the battery sample in the n sliding window, t i is the timestamp, n is the width of the sliding window; when constructing the degradation matrix DM, the discharge capacity-voltage curve of each battery sample in the training sample set is first filtered using linear interpolation so that the discharge signal is regularized into the same shape, and then obtained using sliding window sampling, thereby effectively capturing the capacity degradation of the battery.

[0083] In step S203, the trained lithium battery performance prediction model is fine-tuned using the early cycle data of the target battery.

[0084] In an embodiment of the present invention, early cycle data of the target battery are obtained, which may include charge and discharge curves, capacity decay curves, coulombic efficiency, voltage change data, etc., and corresponding GAF matrix and DM matrix are constructed according to the early cycle data. The GAF matrix and DM matrix of the target battery are input into the offline trained lithium battery performance prediction model, and the learning rate of the fine-tuning training is adjusted to 0.1 times of the offline training process to achieve fine-tuning of the model.

[0085] In step S204, the performance of the target battery is predicted online using the fine-tuned and trained lithium battery performance prediction model to obtain the remaining service life and health status of the target battery.

[0086] In the embodiment of the present invention, the specific implementation of step S204 can refer to the description of step S102 in the first embodiment, which will not be repeated here.

[0087] In an embodiment of the present invention, a sample screening strategy is used to screen a preset battery sample data set to obtain a training sample set, and a pre-constructed lithium battery performance prediction model is offline trained based on the training sample set to learn the battery degradation pattern, and the offline trained lithium battery performance prediction model is fine-tuned based on the early cycle data of the target battery. The fine-tuned lithium battery performance prediction model is used to perform online prediction of the performance of the target battery to obtain the remaining service life and health status of the target battery, thereby achieving the purpose of reducing the amount of input data and balancing the prediction accuracy by using a two-step training strategy of offline training-fine-tuning, and overcoming the problem that the existing technology can only estimate the number of remaining cycles end-to-end, thereby achieving accurate prediction of the lithium battery capacity degradation trajectory and improving the prediction accuracy and prediction effect of lithium battery performance prediction.

[0088] Embodiment three:

[0089] Figure 5 The structure of the lithium battery performance prediction device based on deep learning provided in the third embodiment of the present invention is shown. For the convenience of explanation, only the part related to the embodiment of the present invention is shown, including:

[0090] A model fine-tuning unit 51 is used to fine-tune the trained lithium battery performance prediction model using early cycle data of the target battery;

[0091] The performance prediction unit 52 is used to use the fine-tuned and trained lithium battery performance prediction model to perform online prediction on the performance of the target battery to obtain the remaining service life and health status of the target battery.

[0092] Preferably, for the training of the lithium battery performance prediction model, the lithium battery performance prediction device based on deep learning in the embodiment of the present invention further includes:

[0093] A sample screening unit, used to screen a preset battery sample data set using a sample screening strategy to obtain a training sample set;

[0094] The model training unit is used to perform offline training on a pre-built lithium battery performance prediction model using a training sample set to obtain a trained lithium battery performance prediction model.

[0095] Preferably, the lithium battery performance prediction model includes a dual-channel feature extraction module, a feature aggregation module and a timing processing and prediction module, wherein the dual-channel feature extraction module includes a first channel module and a second channel module, and the first channel module and the second channel module are both embedded with channel attention and spatial attention modules, the feature aggregation module includes a flattening layer and a splicing layer, and the timing processing and prediction module includes a long short-term memory network, two fully connected layers and a timing attention module.

[0096] Preferably, the first channel module and the second channel module each include two convolutional layers, and each convolutional layer is connected to a maximum pooling layer. At the same time, a channel attention and spatial attention module are embedded between each convolutional layer and the maximum pooling layer.

[0097] Preferably, the model training unit includes:

[0098] A matrix construction unit is used to construct a quasi-Gram angle field matrix and a degradation matrix of battery samples in the training sample set according to the training sample set;

[0099] The model training subunit uses the quasi-Gram angular field matrix and degradation matrix to perform offline training on the lithium battery performance prediction model.

[0100] Preferably, the sample screening unit comprises:

[0101] A data cleaning unit, used to clean the battery sample data set according to a preset capacity drop threshold to obtain a remaining sample data set;

[0102] A matrix extraction unit, used for extracting the capacity degradation matrix of all battery samples in the remaining sample data set according to the remaining sample data set;

[0103] A matrix dimension reduction unit, used for performing dimension reduction processing on the capacity degradation matrix by using principal component analysis to obtain a capacity dimension reduction matrix;

[0104] A matrix clustering unit is used to perform clustering processing on the capacity dimension reduction matrix using an unsupervised learning algorithm to obtain a clustering result;

[0105] The sample screening subunit is used to screen the remaining sample data sets based on preset prior features and clustering results to obtain a training sample set.

[0106] In the embodiment of the present invention, each unit of the lithium battery performance prediction device based on deep learning can be implemented by a corresponding hardware or software unit, and each unit can be an independent software or hardware unit, or can be integrated into a software or hardware unit, which is not intended to limit the present invention. Specifically, the implementation of each unit can refer to the description of the aforementioned method embodiment, which will not be repeated here.

[0107] Embodiment 4:

[0108] Figure 6 The structure of an electronic device provided by a fourth embodiment of the present invention is shown. For the convenience of description, only the parts related to the embodiment of the present invention are shown.

[0109] The electronic device 6 of the embodiment of the present invention includes a processor 60, a memory 61, and a computer program 62 stored in the memory 61 and executable on the processor 60. When the processor 60 executes the computer program 62, the steps in the embodiment of the lithium battery performance prediction method based on deep learning are implemented, for example Figure 1 Alternatively, when the processor 60 executes the computer program 62, the functions of each unit in the above-mentioned device embodiments are realized, for example Figure 5 The functions of the units 51 to 52 are shown.

[0110] In an embodiment of the present invention, the early cycle data of the target battery is used to perform fine-tuning training on the trained lithium battery performance prediction model, and the fine-tuned lithium battery performance prediction model is used to perform online prediction of the performance of the target battery to obtain the remaining service life and health status of the target battery, wherein, before fine-tuning the lithium battery performance prediction model, a sample screening strategy is used to screen a preset battery sample data set to obtain a training sample set, and the training sample set is used to perform offline training on the pre-built lithium battery performance prediction model to obtain a trained lithium battery performance prediction model.

[0111] The electronic device of the embodiment of the present invention may be a personal computer or a server. The steps implemented when the processor 60 in the electronic device 6 executes the computer program 62 to implement the lithium battery performance prediction method based on deep learning can refer to the description of the aforementioned method embodiment, which will not be repeated here.

[0112] Embodiment five:

[0113] In an embodiment of the present invention, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps in the above-mentioned lithium battery performance prediction method embodiment based on deep learning are implemented, for example, Figure 1 Alternatively, when the computer program is executed by a processor, the functions of each unit in the above-mentioned device embodiments are realized, for example Figure 5 The functions of the units 51 to 52 are shown.

[0114] In an embodiment of the present invention, the early cycle data of the target battery is used to perform fine-tuning training on the trained lithium battery performance prediction model, and the fine-tuned lithium battery performance prediction model is used to perform online prediction of the performance of the target battery to obtain the remaining service life and health status of the target battery, wherein, before fine-tuning the lithium battery performance prediction model, a sample screening strategy is used to screen a preset battery sample data set to obtain a training sample set, and the training sample set is used to perform offline training on the pre-built lithium battery performance prediction model to obtain a trained lithium battery performance prediction model.

[0115] The computer-readable storage medium of the embodiment of the present invention may include any entity or device or recording medium capable of carrying computer program code, for example, ROM / RAM, magnetic disk, optical disk, flash memory and other memories.

[0116] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A lithium battery performance prediction method based on deep learning, characterized in that: The method comprises the following steps: Use the early cycle data of the target battery to fine-tune the trained lithium battery performance prediction model; Using the fine-tuned and trained lithium battery performance prediction model to perform online prediction on the performance of the target battery, and obtain the remaining service life and health status of the target battery; The lithium battery performance prediction model is trained in the following way: A sample screening strategy is used to screen a preset battery sample data set to obtain a training sample set; The pre-built lithium battery performance prediction model is trained offline using the training sample set to obtain a trained lithium battery performance prediction model.

2. The method according to claim 1, characterized in that The lithium battery performance prediction model includes a dual-channel feature extraction module, a feature aggregation module and a timing processing and prediction module, wherein the dual-channel feature extraction module includes a first channel module and a second channel module, and the first channel module and the second channel module are both embedded with channel attention and spatial attention modules, the feature aggregation module includes a flattening layer and a splicing layer, and the timing processing and prediction module includes a long short-term memory network, two fully connected layers and a timing attention module.

3. The method according to claim 2, characterized in that The first channel module and the second channel module both include two convolutional layers, and each convolutional layer is connected to a maximum pooling layer. At the same time, a channel attention and spatial attention module is embedded between each convolutional layer and the maximum pooling layer.

4. The method according to claim 3, characterized in that The step of using the training sample set to perform offline training on the pre-built lithium battery performance prediction model includes: According to the training sample set, constructing a quasi-Gram angular field matrix and a degradation matrix of battery samples in the training sample set; The lithium battery performance prediction model is trained offline using the quasi-Gram angular field matrix and the degradation matrix.

5. The method according to claim 1, characterized in that The steps of using the sample screening strategy to screen the preset battery sample data set include: Cleaning the battery sample data set according to a preset capacity drop threshold to obtain a remaining sample data set; According to the remaining sample data set, extracting the capacity degradation matrix of all battery samples in the remaining sample data set; Performing dimension reduction processing on the capacity degradation matrix by using principal component analysis to obtain a capacity dimension reduction matrix; Using an unsupervised learning algorithm to perform clustering processing on the capacity dimension reduction matrix to obtain a clustering result; Based on the preset prior features and the clustering results, the remaining sample data sets are screened to obtain the training sample set.

6. A lithium battery performance prediction device based on deep learning, characterized in that: The device comprises: A model fine-tuning unit, used to fine-tune the trained lithium battery performance prediction model using early cycle data of the target battery; A performance prediction unit, used to use the lithium battery performance prediction model that has been fine-tuned and trained to perform online prediction on the performance of the target battery, and obtain the remaining service life and health status of the target battery; The lithium battery performance prediction model is trained in the following way: A sample screening strategy is used to screen a preset battery sample data set to obtain a training sample set; The pre-built lithium battery performance prediction model is trained offline using the training sample set to obtain a trained lithium battery performance prediction model.

7. The device according to claim 6, characterized in that The lithium battery performance prediction model includes a dual-channel feature extraction module, a feature aggregation module and a timing processing and prediction module, wherein the dual-channel feature extraction module includes a first channel module and a second channel module, and the first channel module and the second channel module are both embedded with channel attention and spatial attention modules, the feature aggregation module includes a flattening layer and a splicing layer, and the timing processing and prediction module includes a long short-term memory network, two fully connected layers and a timing attention module.

8. The device according to claim 6, characterized in that The sample screening unit comprises: A data cleaning unit, used for cleaning the battery sample data set according to a preset capacity drop threshold to obtain a remaining sample data set; A matrix extraction unit, configured to extract, according to the remaining sample data set, a capacity degradation matrix of all battery samples in the remaining sample data set; A matrix dimension reduction unit, used for performing dimension reduction processing on the capacity degradation matrix by using principal component analysis to obtain a capacity dimension reduction matrix; A matrix clustering unit, used for performing clustering processing on the capacity dimension reduction matrix using an unsupervised learning algorithm to obtain a clustering result; The sample screening subunit is used to screen the remaining sample data set based on preset prior features and the clustering result to obtain the training sample set.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images