Battery remaining useful life prediction method based on similarity feature enhancement, electronic device and readable storage medium

Abstract

The application provides a battery remaining useful life prediction method based on similarity feature enhancement, which comprises the following steps: S1. Real-time extraction of current data X S and voltage data X V of battery cycle charging and discharging; S2. Determination of current graph G1(X S ,A1) according to the current data X S1 and determination of voltage graph G2(X V ,A2) according to the voltage data X V1 ; S3. Design of a similarity feature enhancement module, input of the current graph G1 and the voltage graph G2 into the similarity feature enhancement module, and obtaining of comprehensive feature H out ; S4. Real-time input of the comprehensive feature H out into a prediction module, and obtaining of the health state of the battery; and battery remaining useful life prediction by using the comprehensive feature obtained by the foregoing method, which can improve the prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of battery life prediction, and more particularly to a method for predicting the remaining battery life based on similarity feature enhancement, an electronic device, and a readable storage medium. Background Technology

[0002] Rechargeable lithium-ion batteries play a crucial role in many modern applications, especially in electric vehicles, where they are widely used primarily due to their advantages such as high energy density, long cycle life, low self-discharge rate, and minimal environmental pollution. As a product of industrialization, lithium-ion batteries exhibit performance degradation. Generally, the predicted remaining useful life (RUL) of a lithium-ion battery can be characterized by its capacity. When the effective capacity of a lithium-ion battery falls below 80% of its initial value, the battery is in an unstable operating state. If the current capacity drops from 80% to 70%, the battery performance deteriorates sharply and is considered a battery failure. Continued operation of the battery may lead to safety accidents, equipment damage, or even battery explosions. Therefore, accurately assessing the RUL of a lithium-ion battery is a prerequisite for its full utilization and safe use. It allows for maximizing battery performance and extending its lifespan, providing subsequent predictive maintenance strategies and energy system management.

[0003] To date, typical methods for predicting the relative lifetime (RUL) of lithium-ion batteries generally fall into two categories: model-based methods and data-driven methods. Model-based methods simplify the internal mechanisms of the battery and then model the battery's capacity degradation to describe its degradation behavior. However, these methods rely heavily on prior knowledge of the system's degradation mechanisms, and in practice, it is difficult to accurately assess the RUL of lithium-ion batteries operating under certain noisy and uncertain environments. Compared to model-based methods, data-driven methods can reduce the time cost of model construction and minimize the interference of improperly constructed models on subsequent lifetime predictions. Current data-driven research methods include artificial neural networks, support vector regression, correlation vector machines, and recurrent neural networks. However, these methods can only handle regular data and cannot effectively extract hidden information and degradation patterns from historical battery data, resulting in shortcomings in both model structure optimization and prediction model accuracy improvement.

[0004] Therefore, in order to solve the above-mentioned technical problems, it is urgent to propose a new technical approach. Summary of the Invention

[0005] In view of this, in order to improve the prediction accuracy of battery remaining lifespan, this invention proposes a battery remaining lifespan prediction method, electronic device, and readable storage medium based on similarity feature enhancement.

[0006] This invention provides a method for predicting the remaining battery life based on similarity feature enhancement, comprising the following steps:

[0007] S1. Real-time extraction of current data X during battery cycle charging and discharging. S and voltage data X V ;

[0008] S2. Based on the current data X S Determine the current diagram G1(X) S1 A1), and according to the voltage data X V Determine the voltage diagram G2(X) V1 A2);

[0009] Among them, X S1 and X V1 These respectively represent the current data X S and the voltage data X V The extracted features are represented by A1 and A2, which represent the current adjacency matrix and the voltage adjacency matrix, respectively.

[0010] S3. Design a similarity feature enhancement module, and input the current graph G1 and the voltage graph G2 into the similarity feature enhancement module to obtain the comprehensive feature H. out ;

[0011] S4. Real-time processing of the comprehensive feature H out The remaining lifespan of the battery is obtained by inputting the data into the prediction module.

[0012] Furthermore, the feature enhancement module obtains the comprehensive feature H through the following steps. out :

[0013] S31. A three-layer graph convolutional layer is used to extract features from the current graph G1 and the voltage graph G2 respectively, to obtain the current features and voltage features of each layer;

[0014] S32. Calculate the current characteristic enhancement coefficient and voltage characteristic enhancement coefficient between each layer, including the current characteristic enhancement coefficient and voltage characteristic enhancement coefficient between the first layer and the second layer, the first layer and the third layer, and the second layer and the third layer;

[0015] S33. Calculate the enhancement feature H between the first layer and the second layer based on the current and voltage characteristics of each layer, as well as the current and voltage enhancement coefficients between the layers. 12 The enhancement feature H between the first and third layers 13 And the enhancement feature H between the second and third layers 23 ;

[0016] S34. Transfer the enhancement feature H between the first layer and the second layer. 12 The enhancement feature H between the first layer and the third layer 13 and the enhancement feature H between the second and third layers 23 By splicing the data, we obtain the comprehensive feature H. out ;

[0017] The comprehensive feature H out The expression is as follows:

[0018] H out =H 12 ||H 13 ||H 23

[0019] Among them, h 12 ||H 13 ||H 23 Indicates the enhancement feature H 12 H 13 and H 23 Then, the parts are assembled.

[0020] Furthermore, the enhanced feature h 12 H 13 and H 23 Calculate using the following formulas respectively:

[0021]

[0022]

[0023]

[0024] Among them, a S12 a S13 and a S23 These represent the current characteristic enhancement coefficients between the first and second layers, the first and third layers, and the second and third layers, respectively; a V12 a V13 and a V23 These represent the voltage characteristic enhancement coefficients between the first layer and the second layer, the first layer and the third layer, and the second layer and the third layer, respectively. This indicates that the first layer of current characteristics is spliced ​​with the second layer of current characteristics. This indicates that the first layer voltage feature and the second layer voltage feature are spliced ​​together; This indicates that the first layer of current characteristics is spliced ​​with the third layer of current characteristics. This indicates that the first layer voltage feature is spliced ​​with the third layer voltage feature; This indicates that the second layer of current characteristics is spliced ​​with the third layer of current characteristics. This indicates that the second layer voltage feature is spliced ​​with the third layer voltage feature.

[0025] Furthermore, the current characteristic enhancement coefficient a S12 a S13 and a S23 Calculated using the following formula:

[0026]

[0027]

[0028]

[0029] Where sigmoid represents the sigmoid function, A S12 A S13 and A S23 Both are current cosine similarity matrices, R S1 R S2 and R S3 L1, L2, and L3 represent the cosine similarity of the current graph in the first, second, and third convolutional layers, respectively, with L1 = 1, L2 = 2, and L3 = 3.

[0030] The voltage characteristic enhancement coefficient a V12 a V13 and a V23 The calculation principle and the current characteristic enhancement coefficient a S12 a S13 and a S23 The calculation principle is the same. During the calculation, the current feature similarity and current cosine similarity matrix in the current feature enhancement coefficient calculation formula are replaced with the corresponding voltage feature similarity and voltage cosine similarity matrix.

[0031] Furthermore, the current cosine similarity matrix A S12 A S13 and A S23 Calculated using the following formula:

[0032]

[0033]

[0034]

[0035] in, and These represent the norms of the first-layer current characteristics, the second-layer current characteristics, and the third-layer current characteristics, respectively.

[0036] The calculation principle of the voltage cosine similarity matrix is ​​the same as that of the current cosine similarity matrix. During the calculation, the current feature and the norm of the current feature in the formula are replaced with the corresponding voltage feature and the norm of the voltage feature.

[0037] Furthermore, the current characteristic and the voltage characteristic are calculated using the following formula:

[0038]

[0039]

[0040] in, This represents the current characteristics output after the current graph undergoes the l-th graph convolution. This represents the voltage characteristics output after the voltage graph undergoes the l-th graph convolution. l = 1, 2, and 3 correspond to three graph convolution layers, respectively. When l = 1... X S1 and X V1 These respectively represent the current data X S and the voltage data X V The extracted features, A1 and A2 represent the current adjacency matrix and the voltage adjacency matrix, respectively. and These represent the normalization operation performed on the adjacency matrix. and represents the weight matrices for current and voltage in layer l-1, respectively, and ReLU represents the activation function.

[0041] Furthermore, the current adjacency matrix A1 and the voltage adjacency matrix A2 are determined by the following method:

[0042] Calculate feature X S1 With current data X S The cosine similarity, and based on feature X S1 With current data X S The cosine similarity is used to construct the current adjacency matrix A1;

[0043] Calculate feature X V1 X with voltage data V The cosine similarity, and based on feature X V1 X with voltage data V The voltage adjacency matrix A2 is constructed using the cosine similarity.

[0044] Furthermore, the feature X S1 and X V1 Calculated using the following formula:

[0045] X S1 =MLP(X S)

[0046] X V1 =MLP(X V )

[0047] Among them, MLP(X) S () indicates that current data X is extracted using a multilayer sensing machine. S Features in MLP(X) V () indicates that voltage data X is extracted using a multilayer sensing machine. V Features of [the text].

[0048] Accordingly, the present invention also proposes an electronic device, comprising:

[0049] The system includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the aforementioned battery remaining life prediction method based on similarity feature enhancement.

[0050] Accordingly, the present invention also proposes a readable storage medium storing computer instructions, which, when executed by a processor, implement the above-described method for predicting remaining battery life based on similarity feature enhancement.

[0051] The beneficial effects of the present invention are as follows: By determining the current map and voltage map and inputting them into the similarity feature enhancement module, the present invention obtains a comprehensive feature that fuses and enhances the current feature and voltage feature. Using the comprehensive feature to predict the remaining battery life can improve the accuracy of the prediction. Attached Figure Description

[0052] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0053] Figure 1 This is a flowchart of the present invention.

[0054] Figure 2 This is a flowchart illustrating the calculation of current and voltage characteristics in this invention.

[0055] Figure 3 This is a flowchart illustrating the calculation of the current characteristic enhancement coefficient and the voltage characteristic enhancement coefficient in this invention.

[0056] Figure 4 This is a flowchart illustrating the calculation of the current cosine similarity matrix and the voltage cosine similarity matrix in this invention. Detailed Implementation

[0057] The present invention will be further described below with reference to the accompanying drawings:

[0058] This invention provides a method for predicting the remaining battery life based on similarity feature enhancement, comprising the following steps:

[0059] S1. Real-time extraction of current data X during battery cycle charging and discharging. S and voltage data X V ;

[0060] S2. Based on the current data X S Determine the current diagram G1(X) S1 A1), and according to the voltage data X V Determine the voltage diagram G2(X) V1 A2);

[0061] Among them, X S1 and X V1 These respectively represent the current data X S and the voltage data X V The extracted features are represented by A1 and A2, which represent the current adjacency matrix and the voltage adjacency matrix, respectively.

[0062] S3. Design a similarity feature enhancement module, and input the current graph G1 and the voltage graph G2 into the similarity feature enhancement module to obtain the comprehensive feature H. out ;

[0063] S4. Real-time processing of the comprehensive feature H out The data is input into the prediction module to obtain the remaining lifespan of the battery. This method improves the accuracy of predicting the remaining battery lifespan.

[0064] In this embodiment, in step S1, the current data X during the battery's cyclic charging and discharging process is extracted in real time. S and voltage data X V The extracted current data X can be processed. S and voltage data X V Preprocessing, such as data cleaning and noise reduction, is performed. Extracting real-time data helps predict the remaining battery life in real time; data preprocessing can eliminate errors, correct outliers, and remove noise. Data cleaning and noise reduction are existing technologies and will not be elaborated upon here.

[0065] In this embodiment, in step S2, based on the current data X S Determine the current diagram G1(X) S1 A1), and according to the voltage data X V Determine the voltage diagram G2(X) V1 A2);

[0066] Among them, X S1 and X V1 These respectively represent the current data X Sand the voltage data X V The extracted features are represented by A1 and A2, which represent the current adjacency matrix and the voltage adjacency matrix, respectively.

[0067] The feature X S1 and X V1 Calculated using the following formula:

[0068] X S1 =MLP(X S )

[0069] X V1 =MLP(X V )

[0070] Among them, MLP(X) S () indicates that current data X is extracted using a multilayer sensing machine. S Features in MLP(X) V () indicates that voltage data X is extracted using a multilayer sensing machine. V Features in;

[0071] Multilayer perceptrons (MLPs) can handle complex nonlinear relationships, thus better capturing complex patterns in data. In power systems, current and voltage signals often contain rich nonlinear features, which are crucial for fault diagnosis and prediction. Directly classifying and identifying raw current and voltage data is very difficult; therefore, feature extraction is needed to simplify the task and improve classification accuracy and efficiency. MLPs can significantly improve the accuracy of classification results by extracting features layer by layer. The process of feature extraction using multilayer perceptrons is an existing technology and will not be elaborated here.

[0072] The current adjacency matrix A1 and the voltage adjacency matrix A2 are determined by the following method:

[0073] Calculate feature X S1 With current data X S The cosine similarity, and based on feature X S1 With current data X S The cosine similarity is used to construct the current adjacency matrix A1;

[0074] Calculate feature X V1 X with voltage data V The cosine similarity, and based on feature X V1 X with voltage data VThe voltage adjacency matrix A2 is constructed using cosine similarity. The calculation of cosine similarity is a prior art technique, and its calculation process will not be elaborated here. Cosine similarity only considers the vector direction and is unaffected by the vector length, making it particularly effective when dealing with sparse vectors. For current (voltage) data, there may be current (voltage) variations of different amplitudes; cosine similarity can ignore these amplitude differences and focus on the direction of current (voltage) changes. When constructing current (voltage) maps using cosine similarity, it has advantages such as normalized metric, insensitivity to vector length, low computational complexity, and high direction sensitivity, effectively helping to analyze and predict current (voltage) behavior.

[0075] In this embodiment, in step S3, a similarity feature enhancement module is designed, and the current graph G1 and the voltage graph G2 are input to the similarity feature enhancement module to obtain the comprehensive feature H. out The feature enhancement module obtains the comprehensive feature H through the following steps. out :

[0076] S31. A three-layer graph convolutional layer is used to extract features from the current graph G1 and the voltage graph G2 respectively, to obtain the current features and voltage features of each layer; for example... Figure 2 As shown;

[0077] The current characteristic and the voltage characteristic are calculated using the following formulas:

[0078]

[0079]

[0080] in, This represents the current characteristics output after the current graph undergoes the l-th graph convolution. This represents the voltage characteristics output after the voltage graph undergoes the l-th graph convolution. l = 1, 2, and 3 correspond to three graph convolution layers, respectively. When l = 1... X S1 and X V1 These respectively represent the current data X S and the voltage data X V The extracted features, A1 and A2 represent the current adjacency matrix and the voltage adjacency matrix, respectively. and These represent the normalization operation performed on the adjacency matrix. and represents the weight matrices for current and voltage in layer l-1, respectively, and ReLU represents the activation function.

[0081] Graph convolution operates directly on the nodes of a graph, fully considering its structural characteristics. This means that when processing current and voltage graphs, graph convolution can better capture the complex relationships and dependencies between nodes, thus extracting more meaningful features. Graph Convolutional Networks (GCNs) can automatically learn and extract low- to high-level features from graph data, which is highly advantageous for feature extraction from current and voltage graphs. Through this capability, GCNs can effectively extract key feature information from complex electrical systems; graph convolutional networks have significant advantages when processing current and voltage graphs. This capability allows GCNs to better adapt to and extract features when processing complex electrical system data.

[0082] S32. Calculate the current characteristic enhancement coefficient and voltage characteristic enhancement coefficient between each layer, including the current characteristic enhancement coefficient and voltage characteristic enhancement coefficient between the first layer and the second layer, the first layer and the third layer, and the second layer and the third layer; for example Figure 3 As shown;

[0083] The current characteristic enhancement coefficient a S12 a S13 and a S23 Calculated using the following formula:

[0084]

[0085]

[0086]

[0087] Where sigmoid represents the sigmoid function, A S12 A S13 and A S23 Both are current cosine similarity matrices, R S1 R S2 and R S3 L1, L2, and L3 represent the cosine similarity of the current graph in the first, second, and third convolutional layers, respectively, with L1 = 1, L2 = 2, and L3 = 3.

[0088] The voltage characteristic enhancement coefficient a V12 a V13 and a V23 The calculation principle and the current characteristic enhancement coefficient a S12 a S13 and a S23 The calculation principle is the same, and the calculation formula is as follows:

[0089]

[0090]

[0091]

[0092] Among them, A V12 A V13 and A V23 Both are voltage cosine similarity matrices, R V1 R V2 and R V3 These represent the cosine feature similarity of the voltage map in the first, second, and third convolutional layers, respectively.

[0093] Furthermore, the calculation process of the current cosine similarity matrix and the voltage cosine similarity matrix is ​​as follows: Figure 4 As shown;

[0094] The current cosine similarity matrix A S12 A S13 and A S23 Calculated using the following formula:

[0095]

[0096]

[0097]

[0098] in, and These represent the norms of the first-layer current characteristics, the second-layer current characteristics, and the third-layer current characteristics, respectively.

[0099] The voltage cosine similarity matrix A V12 A V13 and A V23 The calculation principle is the same as that of the current cosine similarity matrix, and the calculation formula is as follows:

[0100]

[0101]

[0102]

[0103] in, and These represent the norms of the first-layer voltage characteristics, the second-layer voltage characteristics, and the third-layer voltage characteristics, respectively.

[0104] If we directly calculate the cosine similarity between the current map convolution and the voltage map convolution layers, that is, calculate the current cosine similarity matrix A... S12A S13 and A S23 And the voltage cosine similarity matrix A V12 A V13 and A V23 The higher the similarity between features of different layers, the better the corresponding node features have learned the features of their surrounding neighboring nodes, thus generating similarity. However, this ignores the possibility that high similarity may also be due to redundant features that have not interacted with other nodes. To address this issue, cosine similarity is calculated for each layer of graph convolutional features. This allows us to obtain the node connectivity of that layer. Multiplying the cosine similarity difference with the corresponding cosine similarity matrix effectively avoids the possibility that high similarity is caused by redundant features. Specifically, by calculating the feature similarity difference between corresponding layers, a smaller difference indicates that the corresponding node features are redundant. S33. Calculate the enhancement feature H between the first and second layers based on the current and voltage features of each layer, as well as the current and voltage feature enhancement coefficients between layers. 12 The enhancement feature H between the first and third layers 13 And the enhancement feature H between the second and third layers 23 ;

[0105] The enhanced feature H 12 H 13 and H 23 Calculate using the following formulas respectively:

[0106]

[0107]

[0108]

[0109] Among them, a S12 a S13 and a S23 These represent the current characteristic enhancement coefficients between the first and second layers, the first and third layers, and the second and third layers, respectively; a V12 a V13 and a V23 These represent the voltage characteristic enhancement coefficients between the first and second layers, the first and third layers, and the second and third layers, respectively. This indicates that the first layer of current characteristics is spliced ​​with the second layer of current characteristics. This indicates that the first layer voltage feature is spliced ​​with the second layer voltage feature; This indicates that the first layer of current characteristics is spliced ​​with the third layer of current characteristics. This indicates that the first layer voltage feature is spliced ​​with the third layer voltage feature; This indicates that the second layer of current characteristics is spliced ​​with the third layer of current characteristics. This indicates that the voltage features of the second and third layers are concatenated. Feature enhancement is achieved by multiplying the concatenated layer features based on the feature enhancement coefficient.

[0110] S34. The enhancement feature H between the first and second layers... 12 The enhancement feature H between the first and third layers 13 Enhancement feature H between the second and third layers 23 By splicing the data, we obtain the comprehensive feature H. out ;

[0111] The comprehensive feature H out The expression is as follows:

[0112] H out =H 12 ||H 13 ||H 23

[0113] Among them, H 12 ||H 13 ||H 23 Indicates the enhancement feature H 12 H 13 and H 23 The splicing is then performed. The feature enhancement coefficients and splicing features between layers are enhanced using the method described above. Then, the features corresponding to the current and voltage layers are added together to achieve a hierarchical fusion operation of the current and voltage features.

[0114] In this embodiment, in step S4, the comprehensive feature H is integrated in real time. out The data is input into the prediction module to obtain the remaining battery life.

[0115] The prediction module can be a fully connected layer, or a neural network and deep learning model capable of making predictions for various purposes. No specific restrictions are imposed on the prediction module here.

[0116] Accordingly, the present invention also proposes an electronic device, comprising:

[0117] The system includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the aforementioned battery remaining life prediction method based on similarity feature enhancement.

[0118] Accordingly, the present invention also proposes a readable storage medium storing computer instructions, which, when executed by a processor, implement the above-described method for predicting remaining battery life based on similarity feature enhancement.

[0119] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0120] Those skilled in the art will understand that implementing all or part of the processes in the above method embodiments can be accomplished by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchlink, DRAM (SLDRAM), RAMbus, direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A battery remaining useful life prediction method based on similarity feature enhancement, characterized in that: The method comprises the following steps: S1. Extracting current data in real time during the battery cycle charging and discharging process and voltage data ; S2. determining a current map from the current data determining a current map , and determining a voltage map from the voltage data determining a voltage map ; wherein, and respectively represent the features extracted from the current data and the voltage data respectively, and respectively represent the current adjacency matrix and the voltage adjacency matrix; S3. design a similarity feature enhancement module, and input the current map and the voltage map into the similarity feature enhancement module to obtain a comprehensive feature; ; the feature enhancement module obtains the comprehensive feature through the following steps: :​​ S31. using a three-layer graph convolution layer to respectively perform feature extraction on the current graph and the voltage graph to obtain current features and voltage features of each layer; S32. Calculate the current feature enhancement coefficient and the voltage feature enhancement coefficient between each layer, including the current feature enhancement coefficient and the voltage feature enhancement coefficient between the first layer and the second layer, the first layer and the third layer, and the second layer and the third layer; S33. Calculate the enhancement characteristic between the first layer and the second layer from the current characteristic and voltage characteristic of the layers, and the current characteristic enhancement factor and voltage characteristic enhancement factor between the layers , the enhancement characteristic between the first layer and the third layer , and the enhancement characteristic between the second layer and the third layer ; S34. splicing the enhanced features between the first layer and second layer , the enhanced features between the first layer and third layer and the enhanced features between the second layer and third layer to obtain comprehensive features ; The integrated features The expression is as follows: ； wherein representing the enhanced features , and are stitched together; S4. Real-time the integrated features are input into a prediction module to obtain the remaining useful life of the battery. 2.The battery remaining useful lifetime prediction method based on similarity feature enhancement according to claim 1, characterized in that: the enhancement feature , and are calculated by the following equations, respectively: ； ； ； wherein, , and represent the current characteristic enhancement coefficient between the first layer and the second layer, the first layer and the third layer, and the second layer and the third layer, respectively; , and represent the voltage characteristic enhancement coefficient between the first layer and the second layer, the first layer and the third layer, and the second layer and the third layer, respectively; represents that the first layer current characteristic is spliced with the second layer current characteristic, represents that the first layer voltage characteristic is spliced with the second layer voltage characteristic; represents that the first layer current characteristic is spliced with the third layer current characteristic, represents that the first layer voltage characteristic is spliced with the third layer voltage characteristic; represents that the second layer current characteristic is spliced with the third layer current characteristic, represents that the second layer voltage characteristic is spliced with the third layer voltage characteristic. 3.The battery remaining useful lifetime prediction method based on similarity feature enhancement according to claim 2, characterized in that: the current feature enhancement coefficient , and is calculated by the following equation: ； ； ； wherein, represents a function, , and are current cosine similarity matrices, , and represent cosine feature similarities of the current map in the first, second and third layer map convolutional layers, respectively, , and represent the first, second and third layers, respectively, = 1, = 2 and = 3. The voltage feature enhancement coefficient , and The calculation principle of the voltage feature enhancement coefficient is the same as that of the current feature enhancement coefficient , and In calculation, the current feature similarity and the current cosine similarity matrix in the current feature enhancement coefficient calculation formula are replaced by the corresponding voltage feature similarity and voltage cosine similarity matrix. 4.The battery remaining useful lifetime prediction method based on similarity feature enhancement according to claim 3, characterized in that: The current cosine similarity matrix , and is calculated by the following equation: ； ； ； wherein, , and respectively denote the norm of the first, second and third layer current characteristics. The calculation principle of the voltage cosine similarity matrix is the same as that of the current cosine similarity matrix. When calculating, the current feature and the norm of the current feature in the formula are replaced by the corresponding voltage feature and the norm of the voltage feature.

5. The similarity feature enhanced battery remaining useful life prediction method according to any one of claim 2 or claim 4, characterized in that: The current feature and the voltage feature are calculated by the following formula: ； ； wherein, represents the current feature outputted by the lth graph convolution of the current graph, represents the voltage feature outputted by the lth graph convolution of the voltage graph, l = 1, 2, 3 respectively correspond to three layers of graph convolution layer, when l = 1, , , and respectively represent the features extracted according to the current data and the voltage data , and respectively represent the current adjacency matrix and the voltage adjacency matrix, and respectively represent the normalization operation on the adjacency matrix, and respectively represent the weight matrix of the current and voltage in the l-1 layer, and ReLU represents the activation function. 6.The battery remaining useful lifetime prediction method based on similarity feature enhancement according to claim 5, characterized in that: The current adjacency matrix and the voltage adjacency matrix by the method of Computing features Cosine similarity with current data Based on features Cosine similarity with current data Building a current adjacency matrix ; computing features cosine similarity with voltage data cosine similarity with voltage data cosine similarity with voltage data voltage adjacency matrix based on cosine similarity ​ 7.The battery remaining useful lifetime prediction method based on similarity feature enhancement according to claim 6, characterized in that: The features and are calculated by the formula: ； ； where MLP represents extracting features from the current data using a multi-layer perceptron, represents extracting features from the voltage data using a multi-layer perceptron. represents extracting features from the voltage data using a multi-layer perceptron.​ 8. An electronic device, comprising: The method comprises the following steps: A memory and a processor, wherein the memory is used to store a computer program, and the computer program is executed by the processor to realize the battery remaining service life prediction method based on similarity feature enhancement in any one of claims 1-7.

9. A readable storage medium characterized by: The computer instructions are stored in the readable storage medium, and the computer instructions are executed by the processor to realize the battery remaining service life prediction method based on similarity feature enhancement in any one of claims 1-7.

Images