Lithium ion battery health state estimation method considering temperature and charging current

Abstract

The invention relates to the technical field of lithium ion battery health state estimation, and particularly discloses a lithium ion battery health state estimation method considering temperature and charging current, which comprises the following steps: firstly, dividing a constant current charging curve into a plurality of segments, and extracting a capacity increment sequence of each segment as a health feature to characterize battery degradation; then, a parallel hybrid network is developed, and local and global modes in health features are effectively captured by combining the advantages of TCN, Transform and an attention mechanism; according to the method, the adaptability of the SOH of the lithium ion battery under different working temperatures, charging currents and battery chemical properties is improved by adopting meta-learning. According to the invention, the maximum estimation error under the condition of cross temperature and charging current is effectively reduced. When the method is applied to different battery types, the requirement for aging data is reduced by 50%, and the method has robust generalization, high precision and strong practical application potential.

Description

Technical Field

[0001] The present invention relates to the technical field of estimating the state of health of lithium-ion batteries, and specifically to a method for estimating the state of health of lithium-ion batteries considering temperature and charging current. Background Art

[0002] Electric vehicles have advantages such as high energy density, long cycle life, and low self-discharge rate. However, with the increase in the number of charge-discharge cycles, irreversible chemical reactions inside the battery will cause the performance to gradually decline. To quantify this degradation, the state of health (SOH) is usually used as a metric, which is defined as the ratio of the current maximum discharge capacity to the initial capacity. Accurate SOH estimation is crucial for ensuring the safety and reliability of electric vehicles and protecting users and property. Summary of the Invention

[0003] Aiming at the above problems of the prior art, the present invention provides a method for estimating the state of health of lithium-ion batteries considering temperature and charging current, which has robust generalization, high precision, and strong practical application potential, and can efficiently and quickly evaluate the state of health of lithium-ion batteries.

[0004] To achieve the above object, the present invention proposes a method for estimating the state of health of lithium-ion batteries considering temperature and charging current, and the method includes the following steps:

[0005] S1. Obtain data of different charging segments of the lithium-ion battery: Perform cyclic charge-discharge tests on the battery through experiments until the terminal voltage reaches 4.2V and the current drops to the cut-off current of 0.02C, and record the battery charging voltage and capacity data in real time;

[0006] S2. Data processing: Slice the voltage obtained in S1 to obtain the charging capacity sequence corresponding to different voltage sequences, and use the capacity increment as the health feature;

[0007] S3. Build a parallel model of the temporal convolutional network TCN and Transformer to fuse local features and global features: Integrate the parallel neural network TCN-Transformer with meta-learning, and consider the changes in temperature and charging current at the same time;

[0008] S4. Initialize the parameters and perform model pre-training: Apply meta-learning to refine the initialization parameters of the parallel neural network through gradient descent, and use a small amount of data for pre-training;

[0009] S5. TCN-Transformer Parallel Model Training: Use the data obtained in S2 to run the optimized TCN-Transformer parallel model and three deep learning algorithms, namely TCN-Transformer, Transformer, and CNN-LSTM, estimate the SOH under the same conditions, and compare the estimation results.

[0010] S6. Model Validation: Validate the TCN-Transformer parallel model integrating meta-learning under different temperatures and across battery materials.

[0011] Preferably, in S2, the calculation process expression is:

[0012] Q = [Q1, Q2, …, Q i , …, Q n ;

[0013]

[0014] In the formula, Q represents the charging capacity, n represents the number of segments, S represents the voltage step, Floor represents rounding down to the nearest integer, V star represents the starting voltage, V stop represents the termination voltage, and ΔV represents the interval between segment voltages.

[0015] Preferably, in S2, for the accurate estimation of the state of health SOH of lithium-ion batteries based on partial random charging data, divide the given charging capacity sequence Q i into multiple segments, obtain the capacity increment order of each segment, and the calculation formula is:

[0016] Q i = [Q i1 , Q i2 , …, Q im ;

[0017] ΔQ i = [Q i1 , Q i2 , …, Q im - Q i1 ;

[0018] In the formula, m represents the number of segments for further dividing the charging capacity, and ΔQ i represents the capacity increment order of each segment.

[0019] Preferably, in S3, the specific steps for building the parallel model of the temporal convolutional network TCN and Transformer include:

[0020] S311. Extract the capacity increment sequence from all random partial charging segments within each cycle as the health feature for network training; extract the capacity increment sequence from a single random partial charging segment within each cycle;

[0021] S312. Incorporate an embedded attention mechanism in the parallel hybrid network to capture the spatio-temporal features of the health data;

[0022] S313. Conduct pre-training on a large number of training samples in the parallel hybrid network, apply meta-learning, refine the initialization parameters of the parallel network through gradient descent, fine-tune the network using the target dataset, and optimize the meta-model using the limited data available in the target domain;

[0023] S314. Use TCN for one-dimensional time series prediction and preserve the data time order, and given a time series X and the corresponding output Y; where, when the kernel size of TCN is set to 2, the output can be expressed as:

[0024] Y t = W1 * X t + W2 * X t-1 ;

[0025] In the formula, W1 and W2 represent the weights of the convolutional kernels, and the weights are shared among all input sequences;

[0026] S315. Introduce gaps between kernel elements to expand the receptive field of the convolutional process, integrate the residual block connection into TCN, directly add the input data after each layer of convolution to the output, and establish cross-layer connections;

[0027] S316. Adopt a sequence-to-sequence architecture Transformer to eliminate recurrent and convolutional operations, capture long-term dependencies based on dot-product attention and multi-head attention mechanisms, and convert the concatenated result into the required dimension.

[0028] Preferably, in S315, the calculation method of the residual block is as follows:

[0029] Y = f(X + Ψ(X));

[0030] In the formula, Ψ(X) is the input of the residual block, f is the softmax activation function, and Y is the output of the residual block.

[0031] Preferably, in S316, the specific steps for converting the concatenated result into the required dimension are to perform positional encoding on the input data, use functions with different frequencies of sine and cosine to calculate the positional information, and mark the data as relative or absolute data. The calculation formula is:

[0032] PE(pos, 2i) = sin(pos / 10000 2i / dmodel )

[0033] PE(pos, 2i + 1) = cos(pos / 10000 2i / dmodel );

[0034] Wherein, sin represents the sine function, cos represents the cosine function, pos represents the position of the current element, i represents different dimensions, dmodel represents the dimension of the model, and PE(pos, 2i) represents the position encoding calculated at position pos and dimension 2i.

[0035] Preferably, in S316, when capturing long-term dependencies based on dot-product attention, to prevent the dot-product value from becoming too large as the dimension d k increases, and to avoid gradient vanishing and unstable gradients during the attention calculation process, a scaling operation is performed. The calculation formula is:

[0036]

[0037] Wherein, Attention represents that the self-attention mechanism is completed by multiple dot-product operations, Q represents the query vector, K is the key vector, V represents the value vector, softmax represents the activation function, and d k represents the dimension of the key vector.

[0038] Preferably, the multi-head attention mechanism is obtained based on the self-attention mechanism. The calculation formula is:

[0039] Multihead(Q, K, V) = Concat(head1,..., head n )W o ;

[0040] Wherein, head1,..., head n represents the output result of each head, and W o represents a weight matrix used to transform the concatenated result to the dimension required by other parts of the model. Concat represents concatenating the output results on the feature dimension, and Multihead represents the step of concatenating and transforming the outputs of multiple independent attention heads.

[0041] Preferably, TCN is composed of causal convolution, dilated convolution, and residual connection. Transformer is composed of an encoder, a decoder, a self-attention mechanism, position encoding, and a fully connected layer; The TCN branch uses dilated convolution to model local temporal dependencies, and Transformer uses the self-attention mechanism to capture long-range dependencies.

[0042] Preferably, in S3, the specific steps for fusing local and global features are:

[0043] S321. Connect the feature vectors of TCN and Transformer to form a unified representation. The calculation formula is as follows:

[0044]

[0045] In the formula, d T is the output size of the TCN branch, d′ T is the output size of the transformer branch, F TCN represents the TCN neural network, F Transformer represents the Transformer model, F conact represents concatenating the two feature tensors output by the two models according to the dimension of the number of features;

[0046] S322. Integrate the features and apply a linear transformation to project the connected features into a low-dimensional space. The calculation formula is as follows:

[0047] F fused = F concat w fusion + b fusion ;

[0048] In the formula, w fusion and b fusion are the weights and biases of the linear layer;

[0049] S323. Estimate the SOH by passing the fused features through a fully connected layer. The calculation formula is as follows:

[0050] SOH = FC(F fused ) = F fused w out + b out ;

[0051] In the formula, w out and b out are the weights and biases of the output layer.

[0052] Therefore, the present invention proposes a method for estimating the health state of a lithium-ion battery considering temperature and charging current, and its beneficial effects are as follows:

[0053] (1) The TCN-Transformer model constructed by the present invention can quickly adapt to new tasks and achieve high performance with limited data and iterations. This model promotes in-depth learning of the domain knowledge of the source dataset, promotes efficient knowledge transfer, and enhances the flexibility of health feature extraction in real-world applications.

[0054] (2) The parallel structure of the present invention effectively captures local and global dependencies, overcomes the information bottleneck of the serial architecture, reduces latency, improves computational efficiency, and enhances the robustness and adaptability of SOH estimation under different operating conditions.

[0055] (3) The feature extraction and SOH estimation methods proposed by the present invention are superior to other deep learning algorithms and have higher accuracy and robustness in battery aging behavior modeling.

[0056] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0057] Figure 1 It is a schematic diagram of the health features extracted in the first cycle of a method for estimating the state of health of a lithium-ion battery considering temperature and charging current according to the present invention;

[0058] Figure 2 It is a schematic diagram of the results of a method for estimating the state of health of a lithium-ion battery considering temperature and charging current according to the present invention and different SOH estimation algorithms;

[0059] Figure 3 It is a schematic diagram of the error comparison between a method for estimating the state of health of a lithium-ion battery considering temperature and charging current according to the present invention and different SOH estimation algorithms;

[0060] Figure 4 It is a schematic diagram of the flow of a method for estimating the state of health of a lithium-ion battery considering temperature and charging current according to the present invention. Detailed Embodiments

[0061] To make the technical solution, advantages, and objectives of the present invention clearer, the technical solution of the embodiments of the present invention will be described clearly and completely below. The described embodiments are part of the embodiments of the present invention, not all of them. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present invention without creative efforts fall within the protection scope of this application.

[0062] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the ordinary meaning understood by those of ordinary skill in the art to which the present invention pertains.

[0063] As Figures 1 - 4 shown, a method for estimating the state of health of a lithium-ion battery considering temperature and charging current according to an embodiment of the present invention specifically includes the following steps:

[0064] S1. Obtain data of different charging segments of the lithium-ion battery: Conduct cyclic charge and discharge tests on the battery through experiments until the terminal voltage reaches 4.2V and the current drops to the cut-off current of 0.02C, and record the battery charging voltage and capacity data in real time;

[0065] S2. Data processing: Slice the voltage obtained in S1 to obtain the charging capacity sequence corresponding to different voltage sequences, and take the capacity increment as the health feature; The calculation process expression is:

[0066] Q = [Q1, Q2, …, Q i , …, Q n ;

[0067]

[0068] In the formula, Q represents the charging capacity, n represents the number of segments, S represents the voltage step, Floor represents rounding down to the nearest integer, V star represents the starting voltage, V stop represents the termination voltage, and ΔV represents the interval between segment voltages.

[0069] In S2, to achieve an accurate estimation of the state of health SOH of the lithium-ion battery based on partially random charging data, the given charging capacity sequence Q i is divided into multiple segments to obtain the capacity increment order of each segment, and the calculation formula is:

[0070] Q i = [Q i1 , Q i2 , …, Q im ;

[0071] ΔQ i = [Q i1 , Q i2 , …, Q im - Q i1 ;

[0072] In the formula, m represents the number of segments into which the charging capacity is further divided, and ΔQ i represents the capacity increment order of each segment.

[0073] S3. Build a parallel model of the temporal convolutional network TCN and Transformer to fuse local and global features: Integrate the parallel temporal convolutional network TCN-Transformer architecture with meta-learning, and consider the changes in temperature and charging current at the same time;

[0074] The TCN consists of causal convolution, dilated convolution, and residual connections, while the Transformer consists of an encoder, a decoder, self-attention mechanism, positional encoding, and fully connected layers; the TCN branch uses dilated convolution to model local temporal dependencies, and the Transformer uses the self-attention mechanism to capture long-range dependencies.

[0075] In S3, the specific steps to build a parallel model of the Temporal Convolutional Network (TCN) and the Transformer include:

[0076] S311. Extract the capacity increment sequence from all random partial charging segments within each cycle as the health feature for network training; extract the capacity increment sequence from a single random partial charging segment within each cycle, which enhances the flexibility of health feature extraction in real-world applications;

[0077] S312. Incorporate the embedded attention mechanism in the parallel hybrid network to capture the spatio-temporal features of health data;

[0078] S313. Conduct pre-training on a large number of training samples to develop a robust model with optimized parameters. Apply meta-learning to refine the initial parameters of the parallel network through gradient descent, enabling the model to quickly adapt to new tasks and achieve high performance with limited data and iterations. This process promotes in-depth learning of the domain knowledge of the source dataset and facilitates efficient knowledge transfer;

[0079] Fine-tune the network using the target dataset to optimize the meta-model with the limited data available in the target domain. This step enhances the adaptability of the network and ensures accuracy;

[0080] S314. Use the TCN for one-dimensional time series prediction while preserving the data time order, and given a time series X and the corresponding output Y; where, when the kernel size is set to 2, the output can be expressed as:

[0081] Y t = W1 * X t + W2 * X t-1 ;

[0082] In the formula, W1 and W2 represent the weights of the convolutional kernels, and the weights are shared among all input sequences;

[0083] S315. Introduce a gap to expand the receptive field of the dilated convolution process between kernel elements, integrate the residual connection into the TCN, directly add the input data after each layer of convolution to the output, and establish cross-layer connections; this method enables the network to capture dependencies within a longer time range without increasing the number of parameters, alleviates the problem of vanishing gradients, makes the network easier to train, and also enhances the feature learning ability. When dealing with long time series data, it improves performance and convergence speed;

[0084] S316. Eliminate recurrent and convolutional operations using the sequence-to-sequence architecture Transformer, capture long-term dependencies based on dot-product attention and multi-head attention mechanisms, and convert the concatenated result into the required dimension.

[0085] In S315, the calculation method of the residual block is as follows:

[0086] Y = f(X + Ψ(X));

[0087] In the formula, Ψ(X) is the output of the residual block, f is the softmax activation function, and Y is the output of the residual block.

[0088] In S316, the specific steps to convert the concatenated result into the required dimension are to perform positional encoding on the input data, calculate positional information using functions of different frequencies of sine and cosine, mark the data as relative or absolute data, and the calculation formula is:

[0089] PE(pos, 2i) = sin(pos / 10000 2i / dmodel );

[0090] PE(pos, 2i + 1) = cos(pos / 10000 2i / dmodel );

[0091] In the formula, sin represents the sine function, cos represents the cosine function, pos represents the position of the current element, i represents different dimensions, dmodel represents the dimension of the model, and PE(pos, 2i) represents the positional encoding calculated at position pos and dimension 2i.

[0092] In S316, when capturing long-term dependencies based on dot-product attention, to prevent the dot-product value from becoming too large as the dimension d k increases, avoid vanishing gradients and unstable gradients during the attention calculation process, and perform a scaling operation. The calculation formula is:

[0093]

[0094] In the formula, Attention represents the self-attention mechanism completed by multiple dot-product operations, Q represents the query vector, K is the key vector, V represents the value vector, softmax represents the activation function, and d k represents the dimension of the key vector.

[0095] The multi-head attention mechanism is obtained based on the self-attention mechanism. The calculation formula is:

[0096] Multihead(Q, K, V) = Concat(head1,..., head n )Wo ;

[0097] In the formula, head1,..., head n represents the output result of each head, and W o represents a weight matrix used to transform the concatenated result into the dimensions required for other parts of the model. Concat represents concatenating the output results along the feature dimension, and Multihead represents the step of concatenating and transforming the outputs of multiple independent attention heads.

[0098] In S3, the specific steps for fusing local and global features are as follows:

[0099] S321: Connect the feature vectors of TCN and Transformer to form a unified representation. The calculation formula is:

[0100]

[0101] In the formula, d T is the output size of the TCN branch, d′ T is the output size of the transformer branch, F TCN represents the TCN neural network, F Transformer represents the Transformer model, and F conact represents concatenating the two feature tensors output by the two models along the number of features dimension;

[0102] S322: Integrate the features and apply a linear transformation to project the connected features into a low-dimensional space. The calculation formula is:

[0103] F fused = F concat w fusion + b fusion ;

[0104] In the formula, w fusion and b fusion are the weights and biases of the linear layer;

[0105] S323: Estimate the SOH by passing the fused features through a fully connected layer. The calculation formula is:

[0106] SOH = FC(F fused ) = F fused w out + b out ;

[0107] In the formula, w out and b out are the weights and biases of the output layer.

[0108] S4. Initialize parameters and perform model pre-training: Apply meta-learning to refine the initialization parameters of the parallel network through gradient descent, and use a small amount of data for pre-training;

[0109] S5. TCN-Transformer parallel model training: Use the data obtained in step S2 to run the optimized TCN-Transformer parallel model and three deep learning algorithms, namely TCN-Transformer, Transformer, and CNN-LSTM, estimate the SOH under the same conditions, and compare the estimation results;

[0110] S6. Model verification: Verify the model under different temperatures and across battery materials.

[0111] Therefore, the present invention provides a method for estimating the state of health of a lithium-ion battery considering temperature and charging current, which improves the adaptability of the state of health SOH of the lithium-ion battery under different operating temperatures, charging currents, and battery chemistries, and effectively reduces the maximum estimation error under cross-temperature and charging current conditions. When applied to different battery types, the requirement for aging data is reduced by 50%, and it has robust generalization, high precision, and strong potential for practical applications.

[0112] 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 them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements do not make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for estimating the state of health of a lithium-ion battery considering temperature and charging current, characterized in that, The method includes the following steps: S1. Obtain data of different charging segments of a lithium-ion battery: Through experiments, conduct cyclic charge and discharge tests on the battery until the terminal voltage reaches 4.2V and the current drops to a cut-off current of 0.02C, and record the battery charging voltage and capacity data in real time; S2. Data processing: Slice the voltage obtained in S1 to obtain the charging capacity sequence corresponding to different voltage sequences, and use the capacity increment as the health feature; S3. Build a parallel model of the Temporal Convolutional Network (TCN) and Transformer to fuse local features and global features: Integrate the parallel neural network TCN-Transformer with meta-learning, and consider the changes in temperature and charging current simultaneously; S4. Initialize parameters and perform model pre-training: Apply meta-learning to refine the initialization parameters of the parallel neural network through gradient descent, and use a small amount of data for pre-training; S5. Train the TCN-Transformer parallel model: Use the data obtained in S2 to run the optimized TCN-Transformer parallel model and three deep learning algorithms, namely TCN-Transformer, Transformer, and CNN-LSTM, estimate the State of Health (SOH) under the same conditions, and compare the estimation results; S6. Model verification: Verify the TCN-Transformer parallel model integrated with meta-learning under different temperatures and across battery materials; 2. The method for estimating the state of health of a lithium-ion battery considering temperature and charging current according to claim 1, wherein In S2, the calculation process expression is: Q = [Q1, Q2, …, Q i , …, Q n ; Wherein, Q represents the charging capacity, n represents the number of segments, S represents the voltage step, Floor represents rounding down to the nearest integer, V star represents the starting voltage, V stop represents the termination voltage, and ΔV represents the interval between segment voltages.

3. The method for estimating the state of health of a lithium-ion battery considering temperature and charging current according to claim 2, wherein In S2, for the accurate estimation of the state of health (SOH) of a lithium-ion battery based on partial random charging data, a given charging capacity sequence Q i is divided into multiple segments, and the capacity increment order of each segment is obtained. The calculation formula is: Q i = [Q i1 , Q i2 , …, Q im ; ΔQ i = [Q i1 , Q i2 , …, Q im - Q i1 ; where m represents the number of segments into which the charging capacity is further divided, and ΔQ i represents the order of the capacity increment for each segment.

4. A method for estimating the state of health of a lithium-ion battery considering temperature and charging current according to claim 1, characterized in that, In S3, the specific steps for building a parallel model of the Temporal Convolutional Network (TCN) and Transformer include: S311. Extract the capacity increment sequence from all random partial charging segments within each cycle as the health feature for network training; extract the capacity increment sequence from a single random partial charging segment within each cycle; S312. Incorporate an embedded attention mechanism in the parallel hybrid network to capture the spatio-temporal features of the health data; S313. Conduct pre-training on a large number of training samples in the parallel hybrid network, apply meta-learning to refine the initialization parameters of the parallel network through gradient descent, fine-tune the network using the target dataset, and optimize the meta-model using the limited data available in the target domain; S314. Use TCN for one-dimensional time series prediction and preserve the data time order, and given a time series X and the corresponding output Y; where, when the kernel size of TCN is set to 2, the output can be expressed as: Y t = W1 * X t + W2 * X t-1 ; In the formula, W1 and W2 represent the weights of the convolutional kernels, and the weights are shared among all input sequences; S315. Introduce a dilation convolution process between kernel elements to expand the receptive field, integrate residual blocks into TCN, directly add the input data after each layer of convolution to the output, and establish cross-layer connections; S316. Adopt a sequence-to-sequence architecture Transformer to eliminate recurrent and convolutional operations, capture long-term dependencies based on dot-product attention and multi-head attention mechanisms, and convert the concatenated result into the required dimension; 5. The method for estimating the state of health of a lithium-ion battery considering temperature and charging current according to claim 4, wherein In S315, the calculation method of the residual block is as follows: Y = f(X + Ψ(X)); Where, Ψ(X) is the input of the residual block, f is the softmax activation function, and Y is the output of the residual block.

6. The method for estimating the state of health of a lithium-ion battery considering temperature and charging current according to claim 4, characterized in that, In S316, the specific steps to convert the concatenated result into the required dimension are to perform positional encoding on the input data, use functions with different frequencies of sine and cosine to calculate positional information, and mark the data as relative or absolute data. The calculation formula is: PE(pos, 2i) = sin(pos / 10000 2i / dmodel ); PE(pos, 2i + 1) = cos(pos / 10000 2i / dmodel ); Where, sin represents the sine function, cos represents the cosine function, pos represents the position of the current element, i represents different dimensions, dmodel represents the dimension of the model, and PE(pos, 2i) represents the positional encoding calculated at position pos and dimension 2i.

7. A method for estimating the state of health of a lithium-ion battery considering temperature and charging current according to claim 4, characterized in that In S316, when capturing long-term dependencies based on dot-product attention, to prevent the dot-product value from becoming too large as the dimension d k increases, avoid vanishing gradients and unstable gradients during the attention calculation process, and perform a scaling operation. The calculation formula is: Wherein, Attention indicates that the self-attention mechanism is completed by multiple dot product operations, Q represents the query vector, K is the key vector, V represents the value vector, softmax represents the activation function, and d k represents the dimension of the key vector.

8. A method for estimating the state of health of a lithium-ion battery considering temperature and charging current according to claim 7, characterized in that, The multi-head attention mechanism is obtained based on the self-attention mechanism. The calculation formula is: Multihead(Q,K,V) = Concat(head1,...,head n )W o ; where, head1,...,head n represent the output results of each head, and W o represents a weight matrix used to transform the concatenated result to the dimension required by other parts of the model. Concat means concatenating the output results in the feature dimension, and Multihead means the step of concatenating and transforming the outputs of multiple independent attention heads.

9. The method for estimating the state of health of a lithium-ion battery considering temperature and charging current according to claim 1, wherein TCN is composed of causal convolution, dilated convolution, and residual connection. Transformer is composed of an encoder, a decoder, a self-attention mechanism, positional encoding, and a fully connected layer. The TCN branch uses dilated convolution to model local temporal dependencies, and Transformer uses the self-attention mechanism to capture long-range dependencies.

10. The method for estimating the state of health of a lithium-ion battery considering temperature and charging current according to claim 1, wherein In S3, the specific steps to fuse local and global features are: S321. Concatenate the feature vectors of TCN and Transformer to form a unified representation. The calculation formula is: where d T is the output size of the TCN branch, d' T is the output size of the transformer branch, and F TCN represents the TCN neural network, and F Transformer represents the Transformer model, and F conact represents splicing the two feature tensors output by the two models according to the dimension of the number of features; S322. Integrate the features and apply a linear transformation to project the concatenated features into a low-dimensional space. The calculation formula is: F fused = F concat w fusion + b fusion ; where w fusion and b fusion are the weights and biases of the linear layer; S323. Estimate the SOH by passing the fused features through a fully connected layer. The calculation formula is: SOH = FC(F fused ) = F fused w out +b out ; where w out and b out are the weights and biases of the output layer.

Images