Lithium ion battery SOH-RUL joint estimation method based on Dozerformer model

Abstract

The invention belongs to the technical field of batteries based on specific calculation models, and discloses a lithium ion battery SOH-RUL joint estimation method based on a Dozerformer model, which can realize and effectively improve the SOH-RUL joint estimation performance of the battery by the Dozerformer model. The method comprises the following steps: firstly, learning a mapping relation between battery aging characteristics and SOH and RUL by adopting a Dozerformer model, decomposing battery degradation data into a periodic component and a trend component, and introducing a sparse attention mechanism consisting of local sparsity, step length sparsity and change sparsity to learn medium and long-term and short-term characteristics of the periodic component; learning trend component feature information by adopting a linear layer, and combining estimation results of the two to obtain an SOH-RUL joint estimation result; the method comprises the steps that firstly, a Dozerformer model is selected, secondly, a smooth secondary loss function is adopted to guide the Dozerformer model so as to reduce the sensitivity of the model to abnormal values in the training process, and finally, the SOH and the RUL of the lithium ion battery are estimated based on the optimal weight and the offset Dozerformer model, and an SOH-RUL estimation result is output.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of battery technology based on a specific calculation model, and in particular to a lithium-ion battery SOH-RUL joint estimation method based on a Dozerformer model. Background Art

[0002] At present, lithium-ion batteries are widely used in various fields, such as electric vehicles, energy storage devices, and mobile electronic devices, because of their advantages such as convenient cycle charge and discharge, high energy density, and long cycle life. However, as the number of battery charge and discharge times accumulates, its service life will decay, which will affect the reliability and safety of equipment or devices equipped with lithium-ion batteries and pose a potential hazard to users. Battery state of health (SOH) and remaining useful life (RUL), as two important parameters of the battery, describe the degree of battery aging from two different perspectives. Therefore, accurately estimating the SOH and RUL of the battery is crucial to improving the overall performance of the battery. It is generally recommended that the user should replace the battery in time when the battery SOH value drops to 80% or the RUL value is close to 0. The calculation formulas for battery SOH and RUL are as follows: RUL t =rr(2) Wherein, SOHt is the SOH value at the tth charge and discharge cycle, Ct is the maximum capacity at the tth charge and discharge cycle, CN is the nominal capacity of the battery, RULt is the RUL value at the tth charge and discharge cycle, r is the total number of cycles when the battery reaches the end of its life, and rt is the number of t cycles used.

[0003] Many researchers have conducted extensive research on battery SOH and RUL estimation. Currently, estimation methods are mainly divided into two categories: model-based and data-driven methods. Model-based state estimation methods are based on establishing a battery physical model and using mathematical formulas to describe the operation and degradation characteristics of lithium batteries, such as equivalent circuits, electrochemical and empirical models. Data-driven estimation methods do not require the establishment of a battery physical model. The model is trained through observable battery external parameter data (usually voltage, current, temperature, etc.) to map the nonlinear relationship between external parameters and SOH and RUL. Since machine learning models are relatively simple and cannot effectively fit the nonlinear relationship of lithium batteries from a large amount of data, deep learning has gradually attracted the attention of researchers in the field of battery SOH and RUL estimation. Existing technicians have designed a model with a multi-channel parallel structure CNN, LSTM and gated recurrent unit (GRU) with an attention mechanism for SOH and RUL estimation, and verified the model performance on different battery data.

[0004] CNN is limited by the size of the convolution kernel and lacks the ability to capture global sequence information, resulting in poor ability to capture long-term dependencies. Although LSTM or its variant networks have stronger long-term dependency capabilities than CNN, they also suffer from information loss in very long sequences due to their serial structure. The Transformer model can directly establish global dependencies due to its unique self-attention mechanism. Each time step can interact with any other time step, which can easily handle long-term dependencies of sequences. The Transformer model is used to achieve RUL estimation of lithium batteries, but SOH prediction is not achieved. Variational mode decomposition is used to reduce data noise and then the Transformer model is used to extract features to achieve SOH estimation, but the performance requirements of SOH estimation are not met. Although the above estimation methods based on the Transformer model have achieved good prediction results, they do not meet the requirements of simultaneously estimating battery SOH and RUL and do not take into account the influence of periodicity and trend in battery degradation data. Due to the influence of the sensor's own measurement accuracy or other factors, there are outliers in the actual collected data. The models in the above methods are all trained using the mean square error (MSE) loss function. The MSE gradient increases linearly with the prediction error, which causes the model to be very sensitive to outliers. Therefore, a very small number of outliers in the data may cause the overall performance of the model to drop sharply. Summary of the invention

[0005] In order to overcome the above technical defects, an embodiment of the present invention provides a lithium-ion battery SOH-RUL joint estimation method based on the Dozerformer model, and its main technical solutions are as follows:

[0006] S1. The Dozerformer model is used to learn the mapping relationship between battery aging features and SOH and RUL. The battery aging features are decomposed into periodic components and trend components. A sparse attention mechanism consisting of local sparseness, step sparseness and change sparseness is introduced to learn the medium-term and short-term features of the periodic component, and a linear layer is used to learn the feature information of the trend component.

[0007] S2. During model training, the weights and biases of the Dozerformer model are optimized through a smooth quadratic loss function.

[0008] S3. Use the Dozerformer model with optimal weights and biases to estimate the SOH and RUL of lithium-ion batteries and output the SOH-RUL estimation results.

[0009] Furthermore, for an input sequence X, where L is the sequence length and N is the number of features, the main calculation process of the Dozerformer model is as follows:

[0010] S1.1. Decompose the input sequence X into trend component and period component;

[0011] S1.2, calculate the prediction results on the trend component;

[0012] S1.3, embed the periodic component into c feature maps with dimension invariance and divide it into p patches;

[0013] S1.4. Use the Transformer encoder to learn the sequence feature information in the period component, and introduce local sparseness and step-size sparseness strategies in the self-attention mechanism to sparse the attention weights, thereby obtaining the sparse attention mechanism results. Subsequently, the final output of the encoder is generated through the feedforward network layer and residual connection;

[0014] S1.5. Use the Transformer decoder to decode the information output by the encoder through the self-attention mechanism combined with the cross-attention mechanism, and finally obtain the output of the decoder;

[0015] S1.6, use 1×1 convolution to obtain the prediction results on the periodic component;

[0016] S1.7. Add the prediction results on the trend component and the cycle component to get the final result.

[0017] The embodiments of the present invention adopt the above technical solution to achieve the following technical effects:

[0018] 1. The adopted Dozerformer model introduces three attention components to improve the performance of SOH-RUL joint estimation, including local attention to capture short-term dependencies, step attention to learn the periodicity of SOH and RUL, and dynamic extended attention to review historical data points to capture the long-term trends of battery SOH and RUL;

[0019] 2. Use the smooth quadratic loss function to guide the optimization of the weights and biases of the Dozerformer model. Compared with the traditional MSE loss function, the smooth quadratic loss function can effectively reduce the model's sensitivity to data outliers and effectively improve the Dozerformer model's performance in battery SOH-RUL joint estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings exemplarily illustrate the embodiments and constitute a part of the specification, and together with the text description of the specification, are used to explain the exemplary implementation of the embodiments. The embodiments shown are for illustrative purposes only and do not limit the scope of the claims. In all drawings, the same reference numerals refer to similar but not necessarily identical elements.

[0021] Figure 1This is the overall framework diagram for the SOH-RUL joint estimation of lithium-ion batteries in the present invention;

[0022] Figure 2 It is a flow chart of the SOH-RUL joint estimation method of lithium-ion batteries based on the Dozerformer model in the present invention;

[0023] Figure 3 MSE and SQL loss function curves and their derivative curves in the present invention;

[0024] Figure 4 The characteristics of CH10-CH15 and the curves of SOH and RUL changing with the number of cycles;

[0025] Figure 5 SOH and RUL estimation curves of different models on the CH10-CH15 test set;

[0026] Figure 6 SOH and RUL estimation curves for different loss functions on the CH10-CH15 test set. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical scheme 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0028] It should be noted that the descriptions involving "first", "second", etc. in the embodiments of the present invention are only for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0029] In the description of the present invention, it should be understood that the numerical labels before the steps do not identify the order of executing the steps, but are only used to facilitate the description of the present invention and distinguish each step, and therefore cannot be understood as a limitation of the present invention.

[0030] In order to solve the above problems, the present invention studies the Dozerformer model for the joint estimation of SOH and RUL of lithium-ion batteries. In addition, we adopt a novel smooth quadratic loss function to reduce the sensitivity of the model to outliers to guide the adjustment of model weights.

[0031] The overall framework for the combined SOH-RUL estimation of lithium-ion batteries is as follows: Figure 1 As shown. The Dozerformer model is used to learn the mapping relationship between battery aging characteristics and SOH and RUL, and the weights and biases of the Dozerformer model are optimized through the smooth quadratic loss function. Finally, the battery SOH-RUL estimation results are visualized. The Dozerformer model and smooth quadratic loss function are introduced in detail below.

[0032] To facilitate those skilled in the art to understand the technical solutions provided by the embodiments of the present invention, the relevant technologies are described below:

[0033] See also Figure 2 , the lithium-ion battery SOH-RUL joint estimation method based on the Dozerformer model in an embodiment of the present invention includes:

[0034] S1. The Dozerformer model is used to learn the mapping relationship between battery aging characteristics and SOH and RUL. The battery aging characteristics are decomposed into periodic components and trend components. A sparse attention mechanism composed of local sparsity, step sparsity and change sparsity is introduced to learn the medium-term and short-term characteristics of the periodic component, and a linear layer is used to learn the feature information of the trend component.

[0035] For the SOH and RUL estimation tasks, many prediction methods based on LSTM and its variant networks have been established. However, as the sequence length increases, the LSTM and its variant networks may experience gradient explosion or vanishing and the computational efficiency decreases accordingly. In contrast, the Transformer model avoids traditional recursive operations and effectively captures global information by using a self-attention mechanism to parallelly calculate battery degradation features for the entire historical sequence. However, the traditional Transformer model ignores the impact of periodicity and trend in time series data on the battery degradation process. Therefore, this paper proposes the Dozerformer model for lithium battery SOH-RUL joint estimation, which can make full use of the periodicity and trend in the battery degradation process to improve the estimation effect.

[0036] For an input sequence X, where L is the sequence length and N is the number of features, the main calculation process of the Dozerformer model is as follows:

[0037] S1.1. Decompose the input sequence X into trend component and period component. The calculation formula is as follows:

[0038]

[0039] Xs=X-Xt∈RL×N(5),

[0040] Among them, Xt is the trend component, Xs is the period component, Avgpool is the average pooling operation, and Padding is the padding operation.

[0041] S1.2. Calculate the forecast results on the trend component. The calculation formula is as follows:

[0042]

[0043] Where T is the length of the predicted sequence, is the trainable weight matrix, is a trainable bias vector.

[0044] S1.3, the periodic component X s The dimension is unchanged and embedded into c feature maps and divided into p patches. The calculation formula is as follows:

[0045]

[0046] Reshape changes the data dimension so that the data can be embedded in the feature map, Conv2D is a two-dimensional convolution operation with a convolution kernel of 3×1, and Patch is a patch division operation. Independently processing each channel allows the model to better learn the characteristics and patterns of each channel, which can improve the model's expressiveness

[14] . Here, the channel-independent method is also used to input each variable separately into the next model. The patch sequence obtained after channel independence for each feature is shown in the following formula:

[0047]

[0048] S1.4. Use the Transformer encoder to learn the sequence feature information in the periodic component, and introduce local sparse and step-size sparse strategies in the self-attention mechanism to sparse the attention weights, thereby obtaining the sparse attention mechanism result. Subsequently, the final output of the encoder is generated through the feedforward network layer and residual connection.

[0049] The self-attention mechanism in the ordinary Transformer encoder is a multi-head self-attention mechanism, and its calculation process is as follows:

[0050]

[0051] Among them, H is the number of long positions, Wq , W k and W v is the trainable weight matrix, b q , b k and b v is a trainable bias vector. The traditional Transformer model uses a full attention mechanism, that is, learning all the information of the entire sequence. However, as the length of the sequence increases, information redundancy leads to overfitting of the model and a rapid decrease in computational efficiency. The sparse attention mechanism can reduce the amount of computation and reduce information redundancy. Two different methods, local sparseness and step-size sparseness, are used on the self-attention mechanism to sparse the obtained attention weights.

[0052] Specifically, local sparseness is to query the weights on a fixed window, thereby ignoring the rest. For a local window of size w, the local sparse mask matrix can be expressed as:

[0053] Among them, ij represents the element in the i-th row and j-th column of the matrix.

[0054] Step sparse is to query the weights at a fixed step size, and ignore other weights and set them to zero. For a step s sparse mask matrix, it can be expressed as:

[0055]

[0056] The sparse mask matrix on the self-attention mechanism is obtained by merging the local sparse mask matrix and the step sparse mask matrix, as shown in the following formula:

[0057]

[0058] Using formula (17), we can get the result of the sparse attention mechanism as follows:

[0059]

[0060]

[0061] Finally, the output of the encoder is obtained through the feedforward network layer and residual connection:

[0062]

[0063] Among them, ReLU is the activation function and LayerNorm is the layer normalization process.

[0064] S1.5. Use the Transformer decoder to decode the information output by the encoder through the self-attention mechanism combined with the cross-attention mechanism, and finally obtain the output of the decoder.

[0065] Unlike the encoder, the decoder not only uses the self-attention mechanism, but also uses the cross-attention mechanism, where the key and value of the cross-attention mechanism come from the output of the encoder. The decoder calculation process is as follows:

[0066] The same multi-head self-attention mechanism as the Transformer encoder is used to process the input of the decoder (i.e., the local sparse and step-size sparse strategies are introduced in the self-attention mechanism to sparse the attention weights). The output of the multi-head self-attention mechanism is obtained through the calculation process of equations (11)-(20). The cross-attention mechanism not only uses local sparse and step-size sparse, but also uses the variable sparse method. The local sparse matrix, step-size sparse matrix, and variable sparse matrix in the cross-attention mechanism can be expressed as:

[0067]

[0068] Among them, len p is the predicted length, len v Used to specify variable length sequence length.

[0069] The final sparse matrix of the cross attention mechanism is obtained by merging the three sparse matrices:

[0070]

[0071] The output of the encoder is then used as the key and value of the cross-attention mechanism and the input of the decoder is used as the query, and the cross-attention mechanism result is obtained after sparsification:

[0072]

[0073] About Crossattn dec Using equations (21)-(22), we can get the final output of the decoder: That is the final prediction result.

[0074] S1.6, use 1×1 convolution to get the prediction result on the periodic component, and the calculation formula is as follows:

[0075]

[0076] S1.7. Add the prediction results of the trend component and the cycle component to get the final result, and the final result Y is:

[0077] S2. During model training, the weights and biases of the Dozerformer model are optimized through a smooth quadratic loss function.

[0078] At present, mean square error (MSE) and L2 are often used as loss functions in time series prediction. They are very sensitive to outliers in the process of guiding model training, because the gradients of MSE and L2 will increase with the increase of prediction error. In addition, the gradient of L1 loss function is a constant. Although it is not sensitive to outliers, it will cause the model convergence speed to slow down. In order to reduce the sensitivity of the model to outliers during training and maintain a certain convergence speed, this paper uses the SQL function to dynamically adjust the gradient to guide model training. Rational quadratic functions are introduced into the MES loss function to form the SQL loss function. The MSE and SQL loss function curves and their derivative curves with different α values ​​when c=1 are shown in Figure 2. Figure 3 The SQL expression is as follows:

[0079] Among them, α∈[0,1] is the smoothing coefficient, c∈(0,+∞), y a is the true value of the sequence, is the predicted value of the sequence.

[0080] Figure 3 It is the MSE and SQL loss function curve and its derivative curve in the present invention. Figure 3 From the function curve in part a, we can see that compared with the MSE loss function, the SQL loss function is smoother, and its smoothness increases with the increase of the α value. Figure 3 Part b shows that the derivative of the SQL loss function does not increase linearly with the increase of error, but tends to a smaller value. Therefore, it does not have a significant error accumulation effect like the MSE loss function. And as the α value increases, the derivative of SQL is closer to 0, which can more effectively reduce the impact of outliers on the model.

[0081] S3. Use the Dozerformer model with optimal weights and biases to estimate the SOH and RUL of lithium-ion batteries and output the SOH-RUL estimation results.

[0082] The data used for estimation are lithium-ion battery aging data, such as battery internal resistance, average temperature, charging capacity, maximum temperature, discharge capacity, minimum temperature and charging time, etc. The maximum temperature and minimum temperature are both charging temperatures. The above aging characteristic data are standardized and input into the Dozerformer model with optimal weights and biases.

[0083] In the embodiment of the present invention, the Dozerformer model used introduces three attention components to improve the performance of SOH-RUL joint estimation, where local attention is used to capture short-term dependencies, step attention is used to learn the periodicity of SOH and RUL, and dynamic extended attention is used to review historical data points to capture the long-term trends of battery SOH and RUL; at the same time, a smooth quadratic loss function is used to guide the optimization of the weights and biases of the Dozerformer model. Compared with the traditional MSE loss function, the smooth quadratic loss function can effectively reduce the sensitivity of the model to data outliers and effectively improve the Dozerformer model for battery SOH-RUL joint estimation performance.

[0084] Finally, in order to verify the technical effect of the technical solution of the present invention, the effectiveness of the Dozerformer model and the smooth quadratic loss function in the joint estimation of battery SOH-RUL is verified by comparing other advanced Transformer-based time series models and ablation experiments on public data sets. The details are as follows:

[0085] The present invention uses 124 commercial LiFePO graphite battery degradation data sets collected from the Stanford Toyota Research Center of the Massachusetts Institute of Technology. The basic parameters of the LiFePO graphite batteries used are shown in Table 1. In order to explore the impact of different fast charging conditions on service life, all batteries were charged to 80% under 72 current groups of 3.6C-6C, and charged at the remaining constant current and constant voltage levels of the same current. Unlike the diversity of charging, each discharge current is 4C. This paper selects the battery aging data of channels 10-15 in the third batch of data for testing and names them CH10-CH15. And we use the internal resistance, charging time, charging capacity, discharge capacity, minimum temperature, maximum temperature, and average temperature in the data as the input features of the model for training. The above seven features of CH10-CH15 and the curves of SOH and RUL changing with the number of cycles are shown in the figure. Figure 4 As shown above, different features have different orders of magnitude. In order to prevent features with larger values ​​from weakening the influence of features with lower values ​​in the model, we standardize the data set. The standardization formula is as follows: Among them, μ is the data mean and σ is the data standard deviation.

[0086] Table 1: Basic battery parameters

[0087]

[0088] Python 3.10 and PyTorch 2.0 frameworks are used to build and test the model, and the Adam algorithm is used as the adaptive optimizer. The specific parameter settings of the Dozerformer model are shown in Table 2. In order to evaluate the accuracy of SOH and RUL estimation, we use three evaluation indicators: mean absolute error (MAE), root mean square error (RMSE), and determination coefficient (R2) to evaluate the model performance. The three evaluation indicators are expressed as follows:

[0089]

[0090] Among them, T is the number of samples in the sequence, a t is the tth measurement value, The tth estimated value, is the average value.

[0091] Table 2: Dozerformer model parameters

[0092]

[0093]

[0094] The SOH-RUL joint estimation performance of the proposed Dozerformer model on CH10-CH15 was tested. The ratio of model training set, validation set and test set on each battery degradation data is 0.15:0.15:0.7. At the same time, in order to demonstrate the superiority of the Dozerformer model used in the SOH-RUL joint estimation, we compared it with the more advanced prediction models Transformer, Informer, Autoformer, and Fedformer. The parameters of the above four comparison models are shown in Table 3. In this section, all models are trained using the MSE loss function to guide the model. The SOH and RUL estimation curves of Dozerformer and comparison models on the CH10-CH15 test set are shown in Figure 3. Figure 5 As shown, its SOH estimation evaluation indicators are shown in Table 4, and its RUL estimation evaluation indicators are shown in Table 5.

[0095] Table 3: Comparison of model parameter settings

[0096]

[0097] Table 4: SOH estimation evaluation indicators of different models on CH10-CH15

[0098]

[0099]

[0100] Table 5: RUL estimation evaluation indicators of different models on CH10-CH15

[0101]

[0102]

[0103] from Figure 5 From the SOH and RUL estimation results in Tables 4-5, it can be seen that: regardless of the estimation results on SOH or RUL, the estimated values ​​of the Transformer and Informer models are close to a straight line and the evaluation indicators are at a high level. The main reason is that the traditional multi-head self-attention mechanism used in the model cannot effectively learn the trend in the battery degradation path. Compared with the above two models, the Autoformer and Fedformer models introduce the sequence decomposition part to make the estimated curve closer to the true value, but at the end of the battery life, there is a significant deviation from the true value and the predicted value fluctuates greatly. The Dozerformer model used in this article performs best in the SOH-RUL joint estimation. The model decomposes the data into trend components and periodic components, and then effectively learns the degradation characteristics of lithium batteries through three attention components, thereby effectively improving the SOH-RUL joint estimation performance.

[0104] Comparison of experimental results of different loss functions:

[0105] The influence of SQL loss function on the joint estimation of SOH-RUL of Dozerformer model was tested on CH10-CH15. The ratio of training set, validation set and test set of each battery degradation data is still 0.15:0.15:0.7. And the common L1 and MSE loss functions are used to compare with the proposed SQL loss function. It is worth noting that in this section, except for the loss function, other parameters are the same. The parameter settings of Dozerformer model are shown in Table 2. The SOH and RUL estimation curves of three different loss functions on CH10-CH15 test sets are shown in Figure 2. Figure 6 As shown, its SOH estimation evaluation indicators are shown in Table 6, and its RUL estimation evaluation indicators are shown in Table 7.

[0106] Table 6: SOH estimation evaluation indicators of different loss functions on CH10-CH15

[0107]

[0108]

[0109] Table 7: RUL estimation evaluation indicators of different loss functions on CH10-CH15

[0110]

[0111] from Figure 6 From the SOH and RUL estimation results in Tables 6-7, we can see that: (1) In the SOH estimation results, the MSE loss function shows the worst estimation performance, mainly because the MSE is very sensitive to outliers, resulting in overfitting of the model training and the worst prediction performance. The average values ​​of MAE (%), RMSE (%) and R2 of the L1 loss function on six batteries are 0.2042%, 3.098% and 0.9925 respectively; the average values ​​of MAE (%), RMSE (%) and R2 of the SQL loss function on six batteries are 0.204%, 0.392% and 0.9927 respectively. Although the evaluation index of the L1 loss function on CH10, CH13 and CH15 is better than that of the SQL loss function, its average value of the evaluation index on all batteries is worse than that of the SQL loss function, so the SQL loss function has the best overall performance. In summary, the SQL loss function can maintain a certain convergence speed while reducing the sensitivity to outliers during model training.

[0112] Conclusion: Accurate estimation of battery SOH and RUL is of great significance for improving the energy management capabilities of electric vehicles. This paper proposes a Dozerformer model with an SQL loss function for the joint estimation of SOH and RUL of lithium batteries. In order to fully learn the overall trend and degradation law of the battery degradation process, the Dozerformer model decomposes the data into trend components and periodic components, and introduces a sparse attention mechanism consisting of a local sparse matrix, a step sparse matrix, and a change sparse matrix to improve the accuracy of the SOH-RUL joint estimation. In order to reduce the sensitivity of the model to outliers during training, the SQL function is used to dynamically adjust the gradient to guide the Dozerformer model training. This improves the model performance and effectively improves the accuracy of SOH-RUL estimation.

[0113] Simulation experiments were conducted on six different batteries using different models and loss functions: Compared with other advanced Transformer-based prediction models, the Dozerformer model proposed in this paper performs better in both SOH estimation and RUL estimation methods, so the model can better learn the lithium battery degradation process. And the SQL loss function used in this paper has an average value of MAE (%), RMSE (%) and R2 in the six battery SOH estimation, which are 0.204%, 0.392% and 0.9927 respectively; the average value of MAE (%), RMSE (%) and R2 in RUL estimation are 6.7912, 6.7918 and 0.998 respectively, which is the best performance among all the listed loss functions.

[0114] Obviously, those skilled in the art should understand that the modules or steps of the above-mentioned embodiments of the present invention can be implemented by general-purpose computer devices, they can be concentrated on a single computer device, or distributed on a network composed of multiple computer devices, and optionally, they can be implemented by executable program codes of computer devices, so that they can be stored in a storage device and executed by the computer device, and in some cases, the steps shown or described can be executed in a different order from that here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. In this way, the embodiments of the present invention are not limited to any specific combination of hardware and software.

[0115] It should be noted that the above are only preferred embodiments of the present invention, and the patent protection scope of the present invention is not limited thereto. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A lithium-ion battery SOH-RUL joint estimation method based on the Dozerformer model, characterized in that: include: S1. The Dozerformer model is used to learn the mapping relationship between battery aging features and SOH and RUL. The battery aging features are decomposed into periodic components and trend components. A sparse attention mechanism consisting of local sparseness, step sparseness and change sparseness is introduced to learn the medium-term and short-term features of the periodic component, and a linear layer is used to learn the feature information of the trend component. S2. During model training, the weights and biases of the Dozerformer model are optimized through a smooth quadratic loss function. S3. Use the Dozerformer model with optimal weights and biases to estimate the SOH and RUL of lithium-ion batteries and output the SOH-RUL estimation results.

2. The lithium-ion battery SOH-RUL joint estimation method according to claim 1, characterized in that: For an input sequence X, where L is the sequence length and N is the number of features, the main calculation process of the Dozerformer model is as follows: S1.

1. Decompose the input sequence X into trend component and period component; S1.2, calculate the prediction results on the trend component; S1.3, embed the periodic component into c feature maps with dimension invariance and divide it into p patches; S1.

4. Use the Transformer encoder to learn the sequence feature information in the period component, and introduce local sparseness and step-size sparseness strategies in the self-attention mechanism to sparse the attention weights, thereby obtaining the sparse attention mechanism results. Subsequently, the final output of the encoder is generated through the feedforward network layer and residual connection; S1.

5. Use the Transformer decoder to decode the information output by the encoder through the self-attention mechanism combined with the cross-attention mechanism, and finally obtain the output of the decoder; S1.6, use 1×1 convolution to obtain the prediction results on the periodic component; S1.

7. Add the prediction results on the trend component and the cycle component to get the final result.

3. The lithium-ion battery SOH-RUL joint estimation method according to claim 2, characterized in that: Local sparseness is to query the weights on a fixed window, thereby ignoring the rest; for a local window of size w, the local sparse mask matrix can be expressed as: Step sparse is to query the weights at a fixed step size, and ignore other weights and set them to zero; for a step s sparse mask matrix, it can be expressed as: At this time, the local sparse mask matrix and the step sparse mask matrix are combined to obtain the sparse mask matrix on the self-attention mechanism, as shown in the following formula: Using formula (17), the result of the sparse attention mechanism is: The output of the encoder obtained through the feedforward network layer and residual connection is: Among them, ReLU is the activation function and LayerNorm is the layer normalization process.

4. The lithium-ion battery SOH-RUL joint estimation method according to claim 2, characterized in that: In S1.5, the Transformer decoder is used to decode the encoder information through the self-attention mechanism and the cross-attention mechanism, which specifically includes the following steps: S1.5.

1. Use the same masked multi-head self-attention mechanism as the Transformer encoder to process the decoder input and obtain the output result of the multi-head self-attention. S1.5.

2. Use the output of the Transformer encoder as the key and value of the cross-attention mechanism, and use the input of the decoder as the query. Introduce the local sparse matrix, step sparse matrix, and change sparse matrix into the cross-attention mechanism, decode the encoder information, and obtain the output result of the cross-attention mechanism. S1.5.

3. The output results of the multi-head self-attention mechanism are fused with the output results of the cross-attention mechanism through the feedforward network layer and the residual connection to generate the final output of the decoder, that is, the final prediction result.

5. The lithium-ion battery SOH-RUL joint estimation method according to claim 4, characterized in that: The local sparse matrix, step sparse matrix, and change sparse matrix on the cross attention mechanism can be expressed as: The local sparse matrix, the step sparse matrix and the change sparse matrix are combined to obtain the final sparse matrix on the cross attention mechanism: After sparseness, the result of the cross attention mechanism is:

6. The method according to claim 1, characterized in that The expression of the smooth quadratic loss function is: Among them, α∈[0,1] is the smoothing coefficient, c∈(0,+∞), y a is the true value of the sequence, is the predicted value of the sequence.

7. The lithium-ion battery SOH-RUL joint estimation method according to claim 1, characterized in that: The battery aging characteristics are characteristics corresponding to the lithium-ion battery aging data, wherein the lithium-ion battery aging data includes: battery internal resistance, average temperature, charging capacity, charging temperature, discharge capacity and charging time.

Images