Lithium battery health state and residual life prediction method and system

Abstract

The invention discloses a lithium battery health state and residual life prediction method and system. The method comprises the following steps: preprocessing original charging data, eliminating noise and abnormal values, and obtaining normalized data; and performing structured processing on the normalized data in a differential mode to generate structured data. And in combination with a convolution trunk line module, a channel sparse attention residual module, a space sparse attention residual module and a global attention residual module, extracting a global time sequence feature map from the structured data. The feature maps are fused with historical data of the health state of the lithium battery to form a fused feature map, prediction is performed based on the fused feature map, the health state and the residual life of the lithium battery are obtained, through an optimized difference mode, the information amount of data can be enriched, the model can sensitively capture subtle state changes in the continuous cycle period, and meanwhile, the model can be used for predicting the health state of the lithium battery. Potential time sequence features in the charging data can be effectively extracted through the attention residual module, and the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the field of lithium-ion battery state prediction, and more specifically, relates to a lithium battery health state and remaining life prediction method and system. Background Art

[0002] Lithium-ion batteries have the advantages of high energy density, long life, no memory effect, and low self-discharge rate. They are widely used in consumer electronics, electric vehicles, and energy storage systems. However, with the extension of usage time, the performance of lithium-ion batteries will gradually decline, which is manifested as a decrease in capacity and a shortened remaining service life. When reaching the end of life, lithium-ion batteries may cause equipment failure and increase safety risks. Therefore, accurately predicting the aging level of lithium-ion batteries is crucial for battery management systems.

[0003] Since deep learning methods are better at handling complex nonlinear scenarios under multiple charging strategies, they have become a promising method for battery state prediction. However, deep learning methods still have limitations in data preprocessing and model building, and it is difficult for the model to fully extract the inherent time series characteristics in the charging data. On the other hand, the difference between battery charging and discharging data cycles is very small, and the discharge data is affected by different operating conditions, which will lead to distribution differences between the data and the model. For the existing differential method of processing battery data, the battery data of the current cycle is differentiated from the battery data of the first cycle. This means that their prediction ability depends largely on the measurement accuracy of the first cycle, and the first cycle after the difference is all 0, resulting in information loss.

[0004] The invention patent with the prior art publication number CN116522140A proposes a multi-task deep learning model and a lithium battery health status and remaining life prediction and control method, including a bottom shared layer, a top task layer and a gated network; the bottom shared layer includes multiple feature extraction models, each of which is used to extract features from the input data respectively; the top task layer is provided with a specific task model in a one-to-one correspondence with the learning task, and the gated network is set in a one-to-one correspondence with the specific task model, and the gated network uses the input data as input features and outputs weight parameters processed by the Softmax function; all the features extracted by the feature extraction model are weighted by the weight coefficient and assembled to obtain the features as the input of the corresponding specific task model to perform specific learning tasks. This scheme mainly relies on gated recurrent units or long short-term memory networks for feature extraction. The model is relatively simple and it is difficult to fully extract the inherent timing features in the charging data. Summary of the invention

[0005] The present invention aims to overcome the defects in the prior art methods for predicting the health status and remaining life of lithium batteries, especially the problem of information loss caused by differentiating the battery data of the current cycle with the battery data of the first cycle, and the shortcomings of deep learning models in capturing potential timing characteristics in charging data. A method and system for predicting the health status and remaining life of lithium batteries are provided.

[0006] The primary purpose of the present invention is to solve the above technical problems. The technical solution of the present invention is as follows:

[0007] A first aspect of the present invention provides a method for predicting the health status and remaining life of a lithium battery, comprising the following steps:

[0008] Preprocess the original charging data to remove noise and outliers to obtain normalized data;

[0009] Using a preset difference method to perform structured processing on the normalized data to obtain structured data;

[0010] Inputting the structured data into a preset multi-strategy attention regression network for feature extraction to obtain a first task global temporal feature graph and a second task global temporal feature graph, wherein the first task and the second task are different prediction tasks;

[0011] The first task global timing feature diagram and the second task global timing feature diagram are respectively combined with the lithium battery health status historical data and input into a preset predictor to output the lithium battery health status prediction result and the remaining life prediction result.

[0012] Furthermore, the preprocessing comprises the following steps:

[0013] Replace outliers with new values ​​generated by linear interpolation;

[0014] Delete the data that disrupts the monotonic decrease of charging capacity;

[0015] Convert the data from the time dimension to the battery state of charge dimension and use a filter to smooth the data;

[0016] Normalize the smoothed data.

[0017] Furthermore, the structured processing comprises the following steps:

[0018] The normalized data of the first t cycles and the normalized data of the most recent L cycles are used as inputs. For each cycle, except the first cycle, the data of the cycle is differentially calculated with the data of the previous cycle to obtain the differential data;

[0019] The differential data is used to form a multi-dimensional data vector and obtain structured data.

[0020] Furthermore, the preset multi-strategy attention regression network includes a convolutional trunk module, a residual module with channel sparse attention, a residual module with spatial sparse attention, and a residual module with global attention; the feature extraction includes the following steps:

[0021] Input structured data into the convolutional trunk module and output preliminary feature maps;

[0022] The preliminary feature map is input into the residual module with channel sparse attention, and the channel temporal feature map is output;

[0023] Input the channel temporal feature map into the residual module with spatial sparse attention and output the spatial temporal feature map;

[0024] The spatial temporal feature map is input into the residual module with global attention and the global temporal feature map is output.

[0025] Furthermore, the preset predictor includes a preset neural network and three multi-layer perceptrons; the method for predicting the health status prediction result and the remaining life of the lithium battery includes the following steps:

[0026] Input the lithium battery health status historical data into the first multi-layer perceptron to adjust the channel size of the data to match the global time series feature map;

[0027] Channel-joining the lithium battery health status historical data after adjusting the channel size with the first task global time series feature map or the second task global time series feature map to obtain a fusion feature map;

[0028] The fused feature map is passed through the neural network, the second multi-layer perceptron, and the third multi-layer perceptron in sequence to output the lithium battery health status prediction result or the remaining life prediction result.

[0029] A second aspect of the present invention provides a lithium battery health status and remaining life prediction system, which is used to implement the steps of a lithium battery health status and remaining life prediction method, and the system includes:

[0030] A preprocessing module is used to preprocess the original charging data to remove noise and outliers to obtain normalized data;

[0031] A structuring module, used for structuring the smoothed data using a preset difference method to obtain structured data;

[0032] The multi-strategy attention regression network module is used to extract features from structured data to obtain the global temporal feature map of the first task and the global temporal feature map of the second task;

[0033] The prediction module is used to input the first task global timing feature diagram and the second task global timing feature diagram into a preset predictor in combination with the lithium battery health status historical data, and output the lithium battery health status prediction result and the remaining life prediction result.

[0034] Furthermore, the raw charging data is preprocessed using a preprocessing module, including the following steps:

[0035] Replace outliers with new values ​​generated by linear interpolation;

[0036] Delete the data that disrupts the monotonic decrease of charging capacity;

[0037] Convert the data from the time dimension to the battery state of charge dimension and use a filter to smooth the data;

[0038] Normalize the smoothed data.

[0039] Furthermore, the normalized data is structured using a structured module, including the following steps:

[0040] The normalized data of the first t cycles and the normalized data of the most recent L cycles are used as inputs. For each cycle, except the first cycle, the data of the cycle is differentially calculated with the data of the previous cycle to obtain the differential data;

[0041] The differential data is used to form a multi-dimensional data vector and obtain structured data.

[0042] Furthermore, the multi-strategy attention regression network module includes a convolution trunk module, a residual module with channel sparse attention, a residual module with spatial sparse attention, and a residual module with global attention; the method for extracting a global temporal feature map using a preset multi-strategy attention regression network module includes the following steps:

[0043] Input structured data into the convolutional trunk module and output preliminary feature maps;

[0044] The preliminary feature map is input into the residual module with channel sparse attention, and the channel temporal feature map is output;

[0045] Input the channel temporal feature map into the residual module with spatial sparse attention and output the spatial temporal feature map;

[0046] The spatial temporal feature map is input into the residual module with global attention and the global temporal feature map is output.

[0047] Furthermore, the prediction module includes a preset neural network and three multi-layer perceptrons; the method for predicting the health status prediction result and the remaining life of the lithium battery using the prediction module includes the following steps:

[0048] Input the lithium battery health status historical data into the first multi-layer perceptron to adjust the channel size of the data to match the global time series feature map;

[0049] Channel-joining the lithium battery health status historical data after adjusting the channel size with the first task global time series feature map or the second task global time series feature map to obtain a fusion feature map;

[0050] The fused feature map is passed through the preset neural network, the second multi-layer perceptron, and the third multi-layer perceptron in sequence to output the lithium battery health status prediction result or the remaining life prediction result.

[0051] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0052] The differential method proposed in the present invention differentiates the battery data of the current cycle from the battery data of the previous cycle, while retaining the values ​​of the first cycle. The processed structured data has richer information and can enable the model to capture the slight state changes between two consecutive cycles, solving the information loss problem of the existing differential method. In addition, the multi-strategy attention regression network is used to fully capture the potential time series features in the charging data, effectively integrating shallow features and deep features to ensure that shallow features are retained when propagating in the deeper layers of the network, thereby improving the accuracy of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to make the purpose and technical solution of the present invention clearer, the present invention provides the following drawings and descriptions:

[0054] Figure 1 Provide a method flow chart for an embodiment of the present invention;

[0055] Figure 2 A schematic diagram of differential structured data provided by an embodiment of the present invention;

[0056] Figure 3 A schematic diagram of the structure of the method provided by an embodiment of the present invention;

[0057] Figure 4 A network structure diagram of three attention strategies provided by an embodiment of the present invention;

[0058] Figure 5 A schematic diagram of a preprocessing module and a differential structuring module provided in an embodiment of the present invention;

[0059] Figure 6 Schematic diagram of the multi-strategy attention regression network module and prediction module provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0061] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0062] Embodiment 1:

[0063] The present invention provides a method for predicting the health status and remaining life of a lithium battery. Figure 1 The figure shows a flow chart of a lithium battery health status and remaining life prediction method, which uses only charging data to jointly predict the health status (SOH) and remaining life (RUL), including terminal voltage, current and temperature (VIT). These three types of data are widely used in various battery-powered devices and are easy to obtain.

[0064] The specific steps are as follows:

[0065] S1: Preprocess the original charging data to remove noise and outliers to obtain normalized data.

[0066] The specific process is:

[0067] Since the raw charging data usually contains noise, the following data cleaning was implemented. First, outliers larger than 15 times the standard deviation were replaced with new values ​​generated by linear interpolation.

[0068] Delete the data that destroys the monotonic decrease of charging capacity to ensure that the charging capacity decreases monotonically.

[0069] The state of charge (SOC) range of a battery is 0%-100% in any cycle. Therefore, a standardized reference is needed to change the variable describing the VIT curve from time to SOC:

[0070] SOC i =C i / C current ×100% (1)

[0071] Among them, C i represents the charging capacity of the i-th sampling point in the current cycle, C current Indicates the discharge capacity of the battery at the end of the current cycle. In addition, the SOC length is resampled to a commonly used length of 224 using linear interpolation.

[0072] The data were smoothed using a Savitzky-Golay (SG) filter with a window length of 5 and an order of 2.

[0073] Since the units and scales of the smoothed data are different, these data are normalized to ensure that the model converges easily:

[0074] X normal =(X raw -X min ) / (X max -X min ) (2)

[0075] Where X normal is the normalized data, X raw is the smoothed data, X min and X max are the minimum and maximum values ​​of all smoothed data respectively.

[0076] In order to adapt to the deep learning model, data structuring is widely used to convert the VIT curve into a structured image. This paper proposes a new difference method to structure the normalized data, where the input data structure consists of the normalized data of the first 5 cycles and the normalized data of the last L cycles. Except for the data of the first cycle, the data of all other cycles are differentiated from the data of the previous cycle, such as Figure 2 As shown. Among them, is a vector containing VIT data, where i and j represent the cycle number and sampling point respectively.

[0077] S2: Perform structured processing on the normalized data using a preset difference method to obtain structured data.

[0078] The specific process is:

[0079] The normalized data of the first five cycles and the normalized data of the most recent L cycles are used as input. For each cycle, except the first cycle, the data of the cycle is differentially calculated with the data of the previous cycle to obtain the differential data.

[0080] Using the differenced data, we can form a multidimensional data vector and obtain structured data, such as Figure 2 As shown, the structured data processed in this way is more informative and can enable the model to capture tiny state changes between two consecutive cycles, which will be more beneficial for the joint prediction of the health status and remaining life of lithium batteries.

[0081] The method first extracts channel and spatial features from I. Then, it extracts global temporal features for a specific task in each branch in the task-specific feature extraction stage. Finally, the extracted task-specific global temporal features of each branch go through the prediction stage to achieve specific predictions. The method structure is as follows Figure 3 shown.

[0082] S3: Input the structured data into a preset multi-strategy attention regression network for feature extraction to obtain a first task global temporal feature map and a second task global temporal feature map, wherein the first task and the second task are different prediction tasks.

[0083] The specific process is:

[0084] Input the structured data into the convolution trunk module and output the preliminary feature map. Given the structured data after differential processing Where H is set to the resampling length of 224, and W is the total number of cycles. The structured cyclic data I first passes through the convolution trunk to obtain the feature map The expression is as follows:

[0085] F1=BN(Conv2d(ReLU(BN(Conv2d(I))))) (3)

[0086] where BN(·), Conv2d(·), and ReLU(·) represent batch normalization, 3×3 convolution operation, and the rectified linear unit (ReLU) activation function, respectively.

[0087] The preliminary feature map is input into the residual module with channel sparse attention, and the channel temporal feature map is output. The feature map F1 passes through the residual module with CSA, which consists of a convolutional residual block and a CSA+downsampling block to obtain the channel temporal feature map. The expression is as follows:

[0088] F2=D1+C1 (4)

[0089] D1=MP(BN(Conv2d(CSA(F1)))) (5)

[0090] C1=MP(Conv2d(F1)) (6)

[0091] where D1 and C1 are the outputs of the CSA+downsampling block and the convolutional residual block, respectively. MP(·) denotes a 2×1 max pooling operation. CSA(·) is the output of the designed CSA.

[0092] The channel temporal feature map is input into the residual module with spatial sparse attention, and the spatial temporal feature map is output. The channel temporal feature map F2 is obtained by the residual module with SSA to obtain the spatial temporal feature map It has similar operations as those in the above expression, except that SSA is used instead of CSA.

[0093] The spatial temporal feature map is input into the residual module with global attention, and the global temporal feature map is output. The feature map F3 passes through two branches in parallel to obtain the global temporal feature map of the specific task. and They also have similar operations as in the above expressions, except that GA is used instead of CSA and the convolution kernel is set to 5×3.

[0094] The above three attention strategies, channel sparse attention, spatial sparse attention, and global attention have similar structures, but multi-head self-attention (MSA) is different, such as Figure 4 As shown in Figure 2, they are all composed of a deep convolutional layer (DWConv), a layer normalization (LN), two or one MSAs with the same structure, a LN, and a feed-forward network (FFN) consisting of two linear layers. To prevent performance loss, there is a residual connection between the input of the first LN and the input of the second LN.

[0095] The GA functions similarly to traditional MSA by merging the H and W dimensions into one dimension, and then re-segmenting the H and W dimensions after MSA, as shown below:

[0096] MSA(Q,K,V)=Concat(head1,…,head h )W o (7)

[0097] head i =Attention(Q i ,K i ,V i ) (8)

[0098]

[0099] Q=W Q X, K=W K X, V=W V X (10)

[0100] where MSA(·) and Attention(·) denote the multi-head self-attention operation and self-attention operation, respectively. Softmax(·) is the softmax function. T denotes transpose. W, X, h, and d denote the number of trainable weights, input data and attention heads, and the channel dimension size of the input data, respectively.

[0101] If GA is used in the early stages of the network, the large number of tokens that attempt to model global information will result in excessive computational cost of self-attention. Therefore, to solve this problem, two new types of self-attention are designed using the sparsification of self-attention to interact with W and H of the input data from different angles. Specifically, CSA puts the channel C in the last dimension, and the expression is as follows:

[0102]

[0103] MSA CSA (·) indicates the MSA operation of CSA.

[0104] SSA places channel C in the penultimate dimension, and the expression is as follows:

[0105]

[0106] MSA SSA (·) indicates the MSA operation of SSA.

[0107] The three attention strategies extract temporal features from different domains respectively. The convolutional residual block in the module can effectively fuse shallow features and deep features to ensure that shallow features are preserved when propagating to deeper layers of the network. In addition, the optimal execution order of the three attention strategies is designed. Since the voltage, current, and temperature data of the battery are highly correlated with each other, performing channel attention first can reveal this correlation feature from the three channels of the input structured data, while performing GA at the end can avoid a lot of computational costs.

[0108] S4: The first task global timing feature diagram and the second task global timing feature diagram are respectively combined with the lithium battery health status historical data and input into a preset predictor, and the lithium battery health status prediction result and the remaining life prediction result are output.

[0109] The specific process is:

[0110] The multi-layer perceptron (MLP) is used to adjust the channel size of the lithium battery health status historical data to match the global time series feature map. As a key health indicator, the lithium battery health status historical data It is crucial for prediction tasks. Due to the global temporal feature map of a specific task and The channel size is different from the lithium battery health status historical data, so the lithium battery health status historical data X SOH Channel expansion is required through MLP, increasing its channel size from 1 to 4C.

[0111] The lithium battery health status historical data after adjusting the channel size is channel-joined with the global time series feature map of the specific task, and the global time series feature map of the specific task is joined with the lithium battery health status historical data after channel expansion to realize fusion feature maps F6 and F7, as shown in FIG. Figure 3 shown.

[0112] The fused feature graph is processed by a bidirectional long short-term memory network (BiLSTM) and two multi-layer perceptrons to output the health status and remaining life prediction results of the lithium battery. The fused feature graph passes through the BiLSTM and two MLPs in turn to achieve the final prediction. These process expressions are as follows:

[0113]

[0114] F6=Concat[MLP(X SOH ),Flatten(F4)] (18)

[0115] in, Represents the predicted state of health (SOH) of lithium batteries. BiLSTM(·), Concat(·), and Flatten(·) represent BiLSTM, channel concatenation, and flattening operations, respectively. Among them, BiLSTM processes the one-dimensional sequence from two directions, discards irrelevant features through its internal memory mechanism, and outputs the sequence at its endpoints. Predicted remaining life prediction results It is implemented similarly according to the above expression.

[0116] Embodiment 2:

[0117] This embodiment provides a lithium battery health status and remaining life prediction system, which is used to implement a lithium battery health status and remaining life prediction method described in Embodiment 1, including:

[0118] A preprocessing module is used to preprocess the original charging data to remove noise and outliers to obtain normalized data;

[0119] A structuring module, used for structuring the smoothed data using a preset difference method to obtain structured data;

[0120] The multi-strategy attention regression network module is used to extract features from structured data to obtain the global temporal feature map of the first task and the global temporal feature map of the second task;

[0121] The prediction module is used to input the first task global timing feature diagram and the second task global timing feature diagram into a preset predictor in combination with the lithium battery health status historical data, and output the lithium battery health status prediction result and the remaining life prediction result.

[0122] More specifically, the raw charging data is preprocessed using a preprocessing module, such as Figure 5 As shown, the following steps are included:

[0123] Replace outliers with new values ​​generated by linear interpolation;

[0124] Delete the data that disrupts the monotonic decrease of charging capacity;

[0125] Convert the data from the time dimension to the battery state of charge dimension and use a filter to smooth the data;

[0126] Normalize the smoothed data.

[0127] More specifically, the structured module is used to perform structured processing on the normalized data, such as Figure 5 As shown, specifically including:

[0128] The normalized data of the first t cycles and the normalized data of the most recent L cycles are used as inputs. For each cycle, except the first cycle, the data of the cycle is differentially calculated with the data of the previous cycle to obtain the differential data;

[0129] The differential data is used to form a multi-dimensional data vector and obtain structured data.

[0130] More specifically, the multi-strategy attention regression network module includes a convolutional trunk module, a residual module with channel sparse attention, a residual module with spatial sparse attention, and a residual module with global attention, such as Figure 6 As shown, the method of extracting the global temporal feature map using the preset multi-strategy attention regression network module specifically includes:

[0131] Input structured data into the convolutional trunk module and output preliminary feature maps;

[0132] The preliminary feature map is input into the residual module with channel sparse attention, and the channel temporal feature map is output;

[0133] Input the channel temporal feature map into the residual module with spatial sparse attention and output the spatial temporal feature map;

[0134] The spatial temporal feature map is input into the residual module with global attention and the global temporal feature map is output.

[0135] More specifically, the prediction module includes a preset neural network and three multi-layer perceptrons, such as Figure 6 As shown; the method for predicting the health status prediction result and the remaining life of a lithium battery using a prediction module comprises the following steps:

[0136] Input the lithium battery health status historical data into the first multi-layer perceptron to adjust the channel size of the data to match the global time series feature map;

[0137] Channel-joining the lithium battery health status historical data after adjusting the channel size with the first task global time series feature map or the second task global time series feature map to obtain a fusion feature map;

[0138] The fused feature map is passed through the preset neural network, the second multi-layer perceptron, and the third multi-layer perceptron in sequence to output the lithium battery health status prediction result or the remaining life prediction result.

[0139] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for predicting the health status and remaining life of a lithium battery, characterized in that: The steps include: Preprocess the original charging data to remove noise and outliers to obtain normalized data; Using a preset difference method to perform structured processing on the normalized data to obtain structured data; Inputting the structured data into a preset multi-strategy attention regression network for feature extraction to obtain a first task global temporal feature graph and a second task global temporal feature graph, wherein the first task and the second task are different prediction tasks; The first task global timing feature diagram and the second task global timing feature diagram are respectively combined with the lithium battery health status historical data and input into a preset predictor to output the lithium battery health status prediction result and the remaining life prediction result.

2. A method for predicting the health status and remaining life of a lithium battery according to claim 1, characterized in that: The pre-processing comprises the following steps: Replace outliers with new values ​​generated by linear interpolation; Delete the data that disrupts the monotonic decrease of charging capacity; Convert the data from the time dimension to the battery state of charge dimension and use a filter to smooth the data; Normalize the smoothed data.

3. A method for predicting the health status and remaining life of a lithium battery according to claim 1, characterized in that: The structured processing comprises the following steps: The normalized data of the first t cycles and the normalized data of the most recent L cycles are used as inputs. For each cycle, except the first cycle, the data of the cycle is differentially calculated with the data of the previous cycle to obtain the differential data; The differential data is used to form a multi-dimensional data vector and obtain structured data.

4. A method for predicting the health status and remaining life of a lithium battery according to claim 1, characterized in that: The preset multi-strategy attention regression network includes a convolutional trunk module, a residual module with channel sparse attention, a residual module with spatial sparse attention, and a residual module with global attention; The feature extraction comprises the following steps: Input structured data into the convolutional trunk module and output preliminary feature maps; The preliminary feature map is input into the residual module with channel sparse attention, and the channel temporal feature map is output; Input the channel temporal feature map into the residual module with spatial sparse attention and output the spatial temporal feature map; The spatial temporal feature map is input into the residual module with global attention and the global temporal feature map is output.

5. A method for predicting the health status and remaining life of a lithium battery according to claim 1, characterized in that: The preset predictor includes a preset neural network and three multi-layer perceptrons; the method for predicting the health status prediction result and the remaining life of the lithium battery includes the following steps: Input the lithium battery health status historical data into the first multi-layer perceptron to adjust the channel size of the data to match the global time series feature map; Channel-joining the lithium battery health status historical data after adjusting the channel size with the first task global time series feature map or the second task global time series feature map to obtain a fusion feature map; The fused feature map is passed through the neural network, the second multi-layer perceptron, and the third multi-layer perceptron in sequence to output the lithium battery health status prediction result or the remaining life prediction result.

6. A lithium battery health status and remaining life prediction system, used to implement a lithium battery health status and remaining life prediction method according to any one of claims 1 to 5, characterized in that: include: A preprocessing module is used to preprocess the original charging data to remove noise and outliers to obtain normalized data; A structuring module, used for structuring the smoothed data using a preset difference method to obtain structured data; A multi-strategy attention regression network module is used to extract features from structured data to obtain a global temporal feature map of the first task and a global temporal feature map of the second task; The prediction module is used to input the first task global timing feature diagram and the second task global timing feature diagram into a preset predictor in combination with the lithium battery health status historical data, and output the lithium battery health status prediction result and the remaining life prediction result.

7. A lithium battery health status and remaining life prediction system according to claim 6, characterized in that: The raw charging data is preprocessed using a preprocessing module, including the following steps: Replace outliers with new values ​​generated by linear interpolation; Delete the data that disrupts the monotonic decrease of charging capacity; Convert the data from the time dimension to the battery state of charge dimension and use a filter to smooth the data; Normalize the smoothed data.

8. A lithium battery health status and remaining life prediction system according to claim 6, characterized in that: The structured module is used to structure the normalized data, including the following steps: The normalized data of the first t cycles and the normalized data of the most recent L cycles are used as inputs. For each cycle, except the first cycle, the data of the cycle is differentially calculated with the data of the previous cycle to obtain the differential data; The differential data is used to form a multi-dimensional data vector and obtain structured data.

9. A lithium battery health status and remaining life prediction system according to claim 6, characterized in that: The multi-strategy attention regression network module includes a convolutional trunk module, a residual module with channel sparse attention, a residual module with spatial sparse attention, and a residual module with global attention; The method for extracting a global temporal feature map using a preset multi-strategy attention regression network module includes the following steps: Input structured data into the convolutional trunk module and output preliminary feature maps; The preliminary feature map is input into the residual module with channel sparse attention, and the channel temporal feature map is output; Input the channel temporal feature map into the residual module with spatial sparse attention and output the spatial temporal feature map; The spatial temporal feature map is input into the residual module with global attention and the global temporal feature map is output.

10. A lithium battery health status and remaining life prediction system according to claim 6, characterized in that: The prediction module includes a preset neural network and three multi-layer perceptrons; the method for predicting the health status prediction result and the remaining life of the lithium battery using the prediction module includes the following steps: Input the lithium battery health status historical data into the first multi-layer perceptron to adjust the channel size of the data to match the global time series feature map; Channel-joining the lithium battery health status historical data after adjusting the channel size with the first task global time series feature map or the second task global time series feature map to obtain a fusion feature map; The fused feature map is passed through the preset neural network, the second multi-layer perceptron, and the third multi-layer perceptron in sequence to output the lithium battery health status prediction result or the remaining life prediction result.

Images