Battery health state estimation method and device, equipment, storage medium and product

Abstract

The invention discloses a battery health state estimation method and device, equipment, a storage medium and a product, and relates to the technical field of battery health management, and the method comprises the steps: carrying out the preprocessing of a collected retired lithium ion battery sample, and obtaining pulse voltage response data under different battery charging states; inputting the pulse voltage response data into a physical constraint diffusion model to generate pulse voltage response characteristic data; generating a battery health state estimation model according to the pulse voltage response characteristic data and the regression model; and performing health state estimation on the to-be-evaluated battery according to the estimation model. According to the method, the pulse voltage response data is extracted from the multi-dimensional retired lithium ion battery sample, then the battery health state estimation model is generated through the physical constraint diffusion model and the regression model, and then the battery health state is estimated according to the estimation model. According to the method, the SOH of the retired power batteries in different dimensions can be quickly and accurately estimated, and the energy consumption in the estimation process is low.

Description

Technical Field

[0001] This application relates to the technical field of battery health management, and particularly to a method, device, equipment, storage medium and product for estimating the state of health of a battery. Background Art

[0002] In the process of the global transition to low-carbon energy, transportation electrification plays an important role, and lithium-ion batteries are the core to achieve transportation electrification. With the wide application of electric vehicles, the number of retired power batteries has increased sharply. It is expected that by 2050, the total capacity of in-service and scrapped electric vehicle batteries will far exceed expectations. However, the management of retired power batteries faces many problems. Their remaining capacity is of great value, usually exceeding 80% of the rated capacity. But due to the lack of proper treatment, it not only brings economic burdens to manufacturers and users, but also causes a series of environmental and social problems such as resource waste, supply chain risks, and carbon emissions.

[0003] To solve the problems brought by battery retirement, the reuse and material recycling of batteries have become the key. In terms of reuse, retired power batteries can be used in fields such as grid energy storage and residential power supply, but pre-treatments such as consistency screening, capacity ranking, and re-grouping are required according to the state of health (SOH) of the retired power batteries to meet the requirements of different applications. In terms of material recycling, the residual value of the battery is utilized through material extraction or structural repair. However, when formulating a direct material recycling strategy, the SOH of the battery determines the required chemical reagents and the expected lithium supplementation dose. Therefore, in summary, in the process of reuse and material recycling of retired power batteries, obtaining accurate SOH information is crucial.

[0004] Obtaining the SOH of retired power batteries faces many challenges. On the one hand, the existing SOH monitoring and recording are mainly carried out during the use stage of electric vehicles. After the retired power battery is separated from the on-vehicle monitoring unit, it is difficult to obtain available SOH data on site. On the other hand, the traditional methods for obtaining SOH have limitations. SOH testing requires a large amount of time and additional power costs. For example, although the hybrid pulse power characterization test can be used for SOH estimation, a complete test sequence takes more than 12 hours. Therefore, for the large number of retired batteries that have arrived, a fast and low-energy SOH estimation method is urgently needed. Summary of the Invention

[0005] The main purpose of this application is to provide a method, device, equipment, storage medium and product for estimating the state of health of a battery, aiming to solve the technical problems of low efficiency and high energy consumption existing in the existing battery SOH estimation methods.

[0006] To achieve the above purpose, this application proposes a method for estimating the state of health of a battery, and the method includes:

[0007] Collect samples of retired lithium-ion batteries and preprocess the samples of retired lithium-ion batteries to obtain pulse voltage response data corresponding to different battery charge states;

[0008] Input the pulse voltage response data into a physically constrained diffusion model to generate pulse voltage response feature data;

[0009] Generate a battery health state estimation model according to the pulse voltage response feature data and a Transformer regression model;

[0010] Estimate the battery to be evaluated according to the battery health state estimation model to obtain the battery health state corresponding to the battery to be evaluated.

[0011] In one embodiment, the step of inputting the pulse voltage response data into a physically constrained diffusion model to generate pulse voltage response feature data includes:

[0012] Determine the noise data of the pulse voltage response data through forward diffusion;

[0013] Input the noise data into a physically constrained diffusion model for physically constrained denoising to obtain theoretical voltage response data;

[0014] Determine a loss function according to the noise data and the theoretical voltage response data;

[0015] Train the physically constrained diffusion model according to the loss function to obtain a trained model;

[0016] Predict the theoretical voltage response data according to the trained model to obtain predicted noise;

[0017] Perform reverse denoising on the pulse voltage response data according to the predicted noise to generate pulse voltage response feature data.

[0018] In one embodiment, the physically constrained diffusion model includes an input layer, a backbone network, and a physically constrained prediction head;

[0019] The step of inputting the noise data into a physically constrained diffusion model for physically constrained denoising to obtain theoretical voltage response data includes:

[0020] Input the noise data into a physically constrained diffusion model, and determine the time step embedding and conditional embedding corresponding to the noise data through the input layer;

[0021] Perform a sampling operation on the time step embedding and the conditional embedding through the backbone network to obtain sampled data;

[0022] Predict the equivalent circuit model parameters corresponding to the sampled data through the physical constraint prediction head, and determine the physical consistency loss based on the equivalent circuit model parameters and the pulse voltage response data;

[0023] Perform physical constraint denoising on the noise data according to the physical consistency loss to obtain theoretical voltage response data.

[0024] In one embodiment, the backbone network includes a downsampling layer, an intermediate layer, and an upsampling layer. The downsampling layer and the upsampling layer are connected through the intermediate layer. The downsampling layer includes a residual block and a cross-attention block;

[0025] The step of performing a sampling operation on the time step embedding and the conditional embedding through the backbone network to obtain the sampled data includes:

[0026] Concatenate the input of the residual block, the time step embedding, and the conditional embedding through the cross-attention block in the downsampling layer to obtain the concatenated input of the residual block;

[0027] Perform a max pooling operation on the concatenated input of the residual block in the residual block of the downsampling layer to compress the feature map size and obtain the downsampled features;

[0028] Capture the global temporal dependencies in the downsampled features through the intermediate layer to obtain intermediate features;

[0029] Perform a feature map size recovery operation on the intermediate features through the upsampling layer to obtain the sampled data.

[0030] In one embodiment, the step of generating a battery health state estimation model according to the pulse voltage response feature data and the Transformer regression model includes:

[0031] Input the pulse voltage response feature data into the Transformer regression model to generate encoded feature data;

[0032] Perform a global average pooling operation on the encoded feature data to obtain pooled feature data;

[0033] Map the pooled feature data to the battery health state estimation range with a preset accuracy through the activation function in the fully connected layer to generate a battery health state estimation model.

[0034] In one embodiment, the Transformer regression model includes an input embedding layer, a multi-head self-attention layer, a feed-forward network, and a layer normalization layer;

[0035] The step of inputting the pulse voltage response characteristic data into the Transformer regression model to generate encoded characteristic data includes:

[0036] Perform a normalization operation on the pulse voltage response characteristic data, and perform a normalization operation on the normalized data to obtain preprocessed characteristic data;

[0037] Map the preprocessed characteristic data to a high-dimensional space through the input embedding layer to obtain high-dimensional characteristic data;

[0038] Perform splicing and linear transformation operations on the high-dimensional characteristic data through the multi-head self-attention layer to obtain transformed characteristic data;

[0039] Perform non-linear activation on the transformed characteristic data through the feed-forward network to obtain non-linearly activated characteristic data;

[0040] Perform a residual connection operation on the transformed characteristic data and the non-linearly activated characteristic data through the layer normalization layer, and perform a layer normalization operation on the connected characteristic data to obtain encoded characteristic data.

[0041] In addition, to achieve the above object, the present application also proposes a battery health state estimation device, and the battery health state estimation device includes:

[0042] A sample preprocessing module, configured to collect retired lithium-ion battery samples and preprocess the retired lithium-ion battery samples to obtain pulse voltage response data corresponding to different battery charge states;

[0043] A physical constraint diffusion module, configured to input the pulse voltage response data into a physical constraint diffusion model to generate pulse voltage response characteristic data;

[0044] An estimation model generation module, configured to generate a battery health state estimation model according to the pulse voltage response characteristic data and the Transformer regression model;

[0045] A health state estimation module, configured to estimate a battery to be evaluated according to the battery health state estimation model to obtain the battery health state corresponding to the battery to be evaluated.

[0046] In addition, to achieve the above object, the present application also proposes a battery health state estimation device, and the device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the battery health state estimation method as described above.

[0047] In addition, to achieve the above object, the present application further provides a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the battery health state estimation method described above are implemented.

[0048] In addition, to achieve the above object, the present application further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the battery health state estimation method described above are implemented.

[0049] The present application provides a battery health state estimation method. By preprocessing the collected retired lithium-ion battery samples, pulse voltage response data corresponding to different battery charge states is obtained; the pulse voltage response data is input into a physically constrained diffusion model to generate pulse voltage response feature data; a battery health state estimation model is generated based on the pulse voltage response feature data and a Transformer regression model; the battery to be evaluated is estimated according to the battery health state estimation model, and the battery health state corresponding to the battery to be evaluated is obtained. Since the present application extracts pulse voltage response data from multi-dimensional retired lithium-ion battery samples, then generates a battery health state estimation model through a physically constrained diffusion model and a Transformer regression model, and then estimates the battery health state according to the estimation model, it can accurately estimate the SOH of retired power batteries in different dimensions, can quickly estimate the SOH of retired batteries, and has low energy consumption, low cost and more environmental protection in the testing and estimation processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0051] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0052] Figure 1 It is a schematic flowchart provided for Embodiment 1 of the battery health state estimation method of the present application;

[0053] Figure 2 It is a schematic diagram of the pulse test curve of the retired power battery in the battery health state estimation method of the present application;

[0054] Figure 3 It is a schematic diagram of the characteristic voltage value output by the model in the battery health state estimation method of the present application;

[0055] Figure 4 Schematic diagram of the SOH value predicted by the model in the battery health state estimation method of the present application;

[0056] Figure 5 Schematic flow chart provided by the second embodiment of the battery health state estimation method of the present application;

[0057] Figure 6 Schematic flow chart provided by the third embodiment of the battery health state estimation method of the present application;

[0058] Figure 7 Overall schematic flow chart of the battery health state estimation method of the present application;

[0059] Figure 8 Schematic calculation flow chart of the SOH fast low - energy consumption estimation model of the present application;

[0060] Figure 9 Schematic diagram of the module structure of the battery health state estimation device in the embodiment of the present application;

[0061] Figure 10 Schematic diagram of the device structure of the hardware operating environment involved in the battery health state estimation method in the embodiment of the present application.

[0062] The realization of the purpose, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. Detailed implementation manners

[0063] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0064] For a better understanding of the technical solutions of the present application, the following will be described in detail in combination with the drawings of the specification and specific implementation manners.

[0065] The main solution of the embodiment of the present application is: collecting samples of retired lithium - ion batteries, pre - processing the samples of retired lithium - ion batteries to obtain pulse voltage response data corresponding to different battery charge states; inputting the pulse voltage response data into a physically - constrained diffusion model to generate pulse voltage response feature data; generating a battery health state estimation model according to the pulse voltage response feature data and a Transformer regression model; estimating the battery to be evaluated according to the battery health state estimation model to obtain the battery health state corresponding to the battery to be evaluated.

[0066] Since there are many challenges in obtaining the State of Health (SOH) of retired power batteries in the existing technology. On the one hand, the existing SOH monitoring and recording are mainly carried out during the use stage of electric vehicles. After the retired power battery is separated from the on-vehicle monitoring unit, it is difficult to obtain the available SOH data on site. On the other hand, the traditional methods for obtaining SOH have limitations. The SOH test requires a large amount of time and additional power costs. For example, although the Hybrid Pulse Power Characterization (HPPC) test can be used for SOH estimation, the complete test sequence takes more than 12 hours. Therefore, in view of the large number of retired batteries that have arrived, there is an urgent need for a fast and low-energy-consuming SOH estimation method.

[0067] This application provides a solution. By preprocessing the collected retired lithium-ion battery samples, the pulse voltage response data corresponding to different battery charge states is obtained; the pulse voltage response data is input into the physically constrained diffusion model to generate pulse voltage response feature data; according to the pulse voltage response feature data and the Transformer regression model, a battery health state estimation model is generated; according to the battery health state estimation model, the battery to be evaluated is estimated to obtain the battery health state corresponding to the battery to be evaluated. Since this application extracts the pulse voltage response data from multi-dimensional retired lithium-ion battery samples, and then generates a battery health state estimation model through the physically constrained diffusion model and the Transformer regression model, and then estimates the battery health state according to the estimation model, it can accurately estimate the SOH of retired power batteries in different dimensions, can quickly estimate the SOH of retired batteries, and the test and estimation processes have low energy consumption, low cost, and are more environmentally friendly.

[0068] It should be noted that the execution subject of the method in this embodiment can be a computing service device with functions such as battery health state estimation, network communication, and program operation, such as a tablet computer, a personal computer, a mobile phone, etc.; it can also be a battery health state estimation device with the same or similar functions. This embodiment and the following embodiments will be described by taking the battery health state estimation device as an example.

[0069] Based on this, the embodiment of this application provides a method for estimating the battery health state, referring to Figure 1 , Figure 1 is the flowchart of the first embodiment of the battery health state estimation method of this application.

[0070] In this embodiment, the battery health state estimation method includes steps S10 to S40:

[0071] Step S10, collect retired lithium-ion battery samples, and preprocess the retired lithium-ion battery samples to obtain pulse voltage response data corresponding to different battery charge states.

[0072] It should be noted that in this embodiment, retired lithium-ion battery samples can be collected first. The battery samples include 3 types of cathode materials (NMC, LFP, LMO), 3 physical forms (cylindrical, soft pack, square), 4 capacity designs (2.1Ah, 10Ah, 21Ah, 35Ah), and 4 historical usage patterns (laboratory accelerated aging, pure electric driving, hybrid driving, etc.). Each battery is subjected to a pulse test at 5% intervals within the range of 5% - 50% SOC (where SOC is the State of Charge). The random SOC conditions of retired batteries are simulated. Pulse test and feature extraction: The battery voltage response curve is obtained by injecting a short-term pulse current (charging for 10 seconds at a 2C rate, standing for 30 seconds, and discharging for 10 seconds). 7-dimensional features (the voltage values at the turning points where the second derivative of the voltage curve is zero) are extracted, and these features reflect aging characteristics such as internal polarization, ohmic impedance, and diffusion impedance changes in the battery.

[0073] It can be understood that the above-collected retired lithium-ion battery samples can be preprocessed to obtain 7-dimensional pulse voltage response data Input data: Pulse voltage response sequence (7-dimensional features (U1 - U7)), extracted from the turning points of the voltage curve of the pulse test. The curve graph of the pulse test of retired power batteries can be referred to Figure 2 as shown. Condition information is extracted, including SOC (State of Charge): Min - Max normalized to [0, 1]; SOH (State of Health): Min - Max normalized to [0, 1]. Battery attributes are extracted, including cathode material type (NMC / LFP / LMO, one-hot encoding), physical form (category encoding), and historical usage pattern (category encoding). Finally, a condition vector is constructed based on the above information, and the above condition information is concatenated into a unified condition vector and mapped to a high-dimensional space through a fully connected layer:

[0074] c embed = ReLU(c·W c + b c ),

[0075] where are trainable parameters, and d embed = 128.

[0076] Step S20, input the pulse voltage response data into a physically constrained diffusion model to generate pulse voltage response feature data.

[0077] It can be understood that a physical constraint diffusion model can be pre-constructed, and physical constraints can be added to the reverse denoising network (U-Net) to obtain a physical constraint diffusion model. Through physical constraints, pulse voltage response characteristic data can be generated to avoid generating data that violates the electrochemical principle (such as a sudden drop in polarization voltage at low SOC).

[0078] Step S30: Generate a battery health state estimation model according to the pulse voltage response characteristic data and the Transformer regression model.

[0079] It can be understood that the Transformer regression model is trained with the pulse voltage response characteristic data to obtain a battery health state estimation model that can estimate the battery health state.

[0080] Step S40: Estimate the battery to be evaluated according to the battery health state estimation model to obtain the battery health state corresponding to the battery to be evaluated.

[0081] It should be understood that estimating the battery to be evaluated according to the battery health state estimation model obtained by the above training can quickly estimate the SOH of the retired battery. The testing and estimation processes have low energy consumption, low cost, are more environmentally friendly, and can make the estimated SOH more accurate. It can accurately estimate the SOH of retired power batteries with different cathode material types, physical formats, capacities, and historical usage conditions. The characteristic voltage value output by the model can be referred to Figure 3 , and the SOH value predicted by the model can be referred to Figure 4 as shown.

[0082] This embodiment provides a method for estimating the battery health state. The collected retired lithium-ion battery samples are preprocessed to obtain pulse voltage response data corresponding to different battery charge states; the pulse voltage response data is input into a physical constraint diffusion model to generate pulse voltage response characteristic data; a battery health state estimation model is generated according to the pulse voltage response characteristic data and the Transformer regression model; the battery to be evaluated is estimated according to the battery health state estimation model to obtain the battery health state corresponding to the battery to be evaluated. Since this application extracts pulse voltage response data from multi-dimensional retired lithium-ion battery samples, then generates a battery health state estimation model through a physical constraint diffusion model and a Transformer regression model, and then estimates the battery health state according to the estimation model, it can accurately estimate the SOH of retired power batteries in different dimensions, can quickly estimate the SOH of retired batteries, and the testing and estimation processes have low energy consumption, low cost, and are more environmentally friendly.

[0083] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar content as that in the above-mentioned first embodiment can be referred to the above introduction and will not be repeated hereinafter. On this basis, please refer to Figure 5 , step S20, the battery health state estimation method further includes steps S201 to S206:

[0084] Step S201, determine the noise data of the pulse voltage response data through forward diffusion.

[0085] It should be noted that the diffusion model destroys the original data distribution by gradually adding noise and defines a noise schedule to control the noise intensity. For the input pulse voltage response data the noise data x at time step t in the forward process t is calculated as follows:

[0086]

[0087] Where, β s follows a linear or cosine scheduling strategy (β 1 = 10 -4 ).

[0088] Step S202, input the noise data into the physically constrained diffusion model for physically constrained denoising to obtain theoretical voltage response data.

[0089] It can be understood that the reverse denoising network (U-Net) can be used as the basic model, and physical constraints are set in the loss function on this basis to obtain the physically constrained diffusion model, and then the noise data is input into the physically constrained diffusion model for physically constrained denoising to obtain theoretical voltage response data.

[0090] In a feasible implementation manner, the physically constrained diffusion model includes an input layer, a backbone network, and a physically constrained prediction head; step S202 may include steps A10 to A40:

[0091] Step A10, input the noise data into the physically constrained diffusion model, and determine the time step embedding and conditional embedding corresponding to the noise data through the input layer.

[0092] It can be understood that the reverse denoising network adopts a U-Net architecture, and the core modules include an input layer, a backbone network, and a physically constrained prediction head. Among them, the input layer first receives the noise data the noise voltage feature at the current time step. Then the input layer calculates the time step embedding, and maps the time step t into a vector in the form of sinusoidal positional encoding Next, the conditional vector is calculated, including conditional information and conditional embeddings. Among them, the conditional information includes SOC (Min-Max normalized), SOH (Min-Max normalized), and battery attributes concatenated as It is mapped to a high-dimensional space through a fully connected layer to obtain the conditional embedding as follows:

[0093] c embed = ReLU(c·W c + b c ),

[0094] where d embed = 128.

[0095] Step A20, perform a sampling operation on the time step embedding and the conditional embedding through the backbone network to obtain the sampled data.

[0096] It can be understood that the time step embedding and the conditional embedding can be sampled through the backbone network. Specifically, the feature map size can be compressed first through downsampling max pooling, then the global temporal dependencies can be captured, and finally the feature map size can be restored through upsampling to obtain the sampled data.

[0097] In a feasible implementation manner, the backbone network includes a downsampling layer, an intermediate layer, and an upsampling layer. The downsampling layer and the upsampling layer are connected through the intermediate layer. The downsampling layer includes residual blocks and cross-attention blocks; Step A20 may include steps A201 to A204:

[0098] Step A201, splice the input of the residual block, the time step embedding, and the conditional embedding through the cross-attention block in the downsampling layer to obtain the input of the spliced residual block.

[0099] It is worth noting that the backbone network includes a downsampling layer, an intermediate layer, and an upsampling layer. The downsampling layer is composed of multiple residual blocks (ResBlock) and attention blocks (AttnBlock), gradually compressing the feature dimensions. Among them, cross-attention inserts a cross-attention module in each downsampling and upsampling layer of the U-Net, using the conditional embedding c embed as the Key and Value, and the current feature as the Query, dynamically adjusting the generation process. In the input of the residual block, the conditional embedding and the time step embedding are spliced through feature concatenation (Concatenation):

[0100] h i ′ n = Concat(h in , c embed , tembed )

[0101] Among them, h i ′ n represents the input of the concatenated residual block.

[0102] Step A202: Input the concatenated residual block into the residual block in the downsampling layer to perform a maximum pooling operation to compress the feature map size, and obtain the downsampled features.

[0103] It should be noted that the residual block (ResBlock) in the downsampling layer:

[0104] h out = GroupNorm(h’ in )W 1 + h’ in ,

[0105] Among them, each residual block is followed by a Swish activation function, h′ in represents the input of the concatenated residual block, and h out represents the output of the residual block.

[0106] It can be understood that the last attention block (AttnBlock) uses a cross-attention mechanism to fuse conditional information:

[0107]

[0108] Among them, d in is the dimension of the input feature, d k is the projection dimension of the query / key, d cond is the dimension of the conditional embedding, and d V is the projection dimension of the value. The downsampling layer compresses the feature map size through maximum pooling.

[0109] Step A203: Capture the global temporal dependencies in the downsampled features through the intermediate layer to obtain intermediate features.

[0110] It can be understood that the intermediate layer is a bottleneck layer connecting the downsampling and upsampling, and contains two ResBlocks and a self-attention block (Self-AttnBlock) for capturing global temporal dependencies to obtain intermediate features.

[0111] Step A204: Perform a feature map size restoration operation on the intermediate features through the upsampling layer to obtain the sampled data.

[0112] It should be understood that the upsampling layer restores the size of the feature map through transposed convolution to obtain the sampled data. The structure of the upsampling layer is symmetric to that of the downsampling layer. Each upsampling block contains a ResBlock and an AttnBlock, and the number of blocks and the connection order are the same as those in the downsampling layer structure.

[0113] Step A30: Predict the equivalent circuit model parameters corresponding to the sampled data through the physical constraint prediction head, and determine the physical consistency loss based on the equivalent circuit model parameters and the pulse voltage response data.

[0114] It should be noted that a lightweight fully connected network is added after the output layer of U-Net as the physical constraint prediction head corresponding to the Equivalent Circuit Model (ECM). The equivalent circuit model parameters (ECM parameters, R 0 , R 1 , C 1 ) corresponding to the output features are predicted through this prediction head:

[0115]

[0116] Among them, h out is the output feature of the last layer of U-Net.

[0117] It can be understood that the theoretical voltage response V ECM (t) can be calculated based on the ECM parameters, and compared with the generated data to calculate the physical consistency loss:

[0118]

[0119] Step A40: Perform physical constraint denoising on the noise data according to the physical consistency loss to obtain the theoretical voltage response data.

[0120] Step S203: Determine the loss function according to the noise data and the theoretical voltage response data.

[0121] It can be understood that the diffusion loss can be calculated based on the noise data, and the mean square error between the predicted noise and the real noise is minimized:

[0122]

[0123] The physical loss can be calculated based on the theoretical voltage response data to constrain the generated voltage curve to conform to the ECM prediction:

[0124]

[0125] Then determine the total loss function: (λ is the balance coefficient and can take a value of 0.1).

[0126] Step S204: Train the physical constraint diffusion model according to the loss function to obtain a trained model.

[0127] It can be understood that the physical constraint diffusion model can be trained according to the loss function, and the AdamW optimizer is used for training, which can be specifically set to a learning rate of 3×10 -4 , and a weight decay of 10 -5 . Batch training can be performed during training, with the batch size set to 64, and mixed-precision training (FP16) can be used to accelerate the calculation. An early stopping mechanism can also be set to monitor the loss of the validation set, and if there is no improvement for 10 consecutive rounds, the training will be terminated.

[0128] Step S205: Predict the theoretical voltage response data according to the trained model to obtain predicted noise.

[0129] Step S206: Perform reverse denoising on the pulse voltage response data according to the predicted noise to generate pulse voltage response feature data.

[0130] It should be noted that the generated data can be verified during the training process. It can be evaluated through the reconstruction error. The mean absolute percentage error (Mean Absolute Percentage Error, MAPE) is used to calculate the mean absolute percentage error between the generated data and the real data (target <0.5%); or the KL divergence (Kullback-Leibler Divergence) can be used to compare the distribution consistency between the generated data and the real data. Then physical consistency checks can be performed to verify the polarization characteristics. At low SOC (such as 5%), the generated data should show an obvious voltage drop (in line with the ECM prediction); and parameter interpretability checks can be performed to check whether the ECM parameters (R 0 , R 1 , C 1 ) change reasonably as the SOH degrades.

[0131] It can be understood that noise prediction is performed at the output layer, and the U-Net finally outputs the predicted noise for reverse denoising. Finally, through DDIM (Denoising Diffusion Implicit Models) sampling and iterative denoising, the finally generated pulse voltage response feature data is obtained

[0132] In this embodiment, it is disclosed to determine the noise data of the pulse voltage response data through forward diffusion, input the noise data into a physically constrained diffusion model for physically constrained denoising to obtain theoretical voltage response data, determine a loss function based on the noise data and the theoretical voltage response data, train the physically constrained diffusion model according to the loss function to obtain a trained model, predict the theoretical voltage response data according to the trained model to obtain predicted noise, and perform reverse denoising on the pulse voltage response data according to the predicted noise to generate pulse voltage response feature data. By using a physically constrained diffusion model to generate the fused pulse voltage response feature data, through the combination of forward diffusion and physical constraints, the diffusion model performs excellently in the generation of complex time-series data, and the details of the generated pulse voltage curve are closer to the real distribution; through ECM constraints, data that violates the electrochemical principle (such as a sudden drop in polarization voltage without low SOC) is avoided, so that more accurate voltage response feature data can be generated.

[0133] Based on the first embodiment of the present application, in the third embodiment of the present application, the same or similar content as in the above-mentioned first embodiment can be referred to the above introduction and will not be repeated hereinafter. On this basis, please refer to Figure 6 , step S30, the battery health state estimation method further includes steps S301 to S303:

[0134] Step S301, input the pulse voltage response feature data into a Transformer regression model to generate encoded feature data.

[0135] In a feasible implementation manner, the Transformer regression model includes an input embedding layer, a multi-head self-attention layer, a feed-forward network, and a layer normalization layer; step S301 may include steps S3011 to S3015:

[0136] Step S3011, perform a normalization operation on the pulse voltage response feature data, and perform a normalization operation on the normalized data to obtain preprocessed feature data.

[0137] It can be understood that the generated 7-dimensional pulse voltage response feature data can be input into the Transformer regression model for encoding processing. First, perform Z-score normalization on each feature dimension:

[0138]

[0139] where μ i and σ i are respectively the mean and standard deviation of the i-th dimension feature of the training set; then perform normalization to obtain the target value corresponding to the encoded feature data: SOH label Normalize to [0, 1].

[0140] Step S3012: Map the preprocessed feature data to a high-dimensional space through the input embedding layer to obtain high-dimensional feature data.

[0141] It should be noted that the 7-dimensional preprocessed feature data is mapped to a high-dimensional space (64-dimensional) through the input embedding layer to obtain high-dimensional feature data:

[0142]

[0143] Step S3013: Perform splicing and linear transformation operations on the high-dimensional feature data through the multi-head self-attention layer to obtain transformed feature data.

[0144] It can be understood that the multi-head self-attention layer performs splicing and linear transformation operations on the high-dimensional feature data:

[0145]

[0146] Let the number of heads h = 4, and the dimension of each head Q = H embed ·W Q K = H embed ·W K V = H embed ·W V W Q W K W V are weight matrices, and the output is obtained by splicing and linear transformation to obtain transformed feature data:

[0147]

[0148] Step S3014: Perform non-linear activation on the transformed feature data through the feed-forward network to obtain non-linearly activated feature data.

[0149] It should be understood that non-linear activation is performed on the transformed feature data through the feed-forward network (FFN):

[0150] H ffn = ReLU(H attn ·W 1 + b 1 )·W 2 + b 2 ,

[0151] where The feedforward network adds non-linear characteristics to the model by introducing non-linear activation functions (such as ReLU, etc.), enabling it to learn more complex patterns and features and enhancing the overall expressive power of the model.

[0152] Step S3015: Perform a residual connection operation on the transformed feature data and the non-linearly activated feature data through the layer normalization layer, and perform a layer normalization operation on the connected feature data to obtain the encoded feature data.

[0153] It should be understood that by performing a residual connection operation on the transformed feature data and the non-linearly activated feature data through the layer normalization layer, and performing a layer normalization operation on the connected feature data, the encoded feature data is obtained:

[0154] H out =LayerNorm(H attn +H ffn ).

[0155] Step S302: Perform a global average pooling operation on the encoded feature data to obtain the pooled feature data.

[0156] It can be understood that by performing a global average pooling operation on the encoded feature data, the sequence dimension 7 is compressed into 1 dimension to obtain the pooled feature data:

[0157]

[0158] Step S303: Map the pooled feature data to the battery health state estimation range with a preset accuracy through the activation function in the fully connected layer to generate a battery health state estimation model.

[0159] It should be understood that by mapping the pooled feature data to the battery health state estimation range with a preset accuracy through the activation function in the fully connected layer:

[0160]

[0161] where σ is the Sigmoid function, which limits the output to [0, 1] and then anti-normalizes it to the original SOH range.

[0162] It can be understood that the model can be trained, and the loss function can be set to mean squared error

[0163]

[0164] Train using the AdamW optimizer, specifically, it can be set to a learning rate of 3×10 -4 , and a weight decay of 10 -4Set the learning rate schedule: Cosine annealing. Set the early stopping mechanism: If the validation loss has not improved for 10 consecutive rounds, terminate the training early, and finally obtain the battery health state estimation model.

[0165] In this embodiment, it is disclosed that the pulse voltage response feature data is input into the Transformer regression model to generate encoded feature data. A global average pooling operation is performed on the encoded feature data to obtain pooled feature data. The pooled feature data is mapped to the battery health state estimation range with a preset accuracy through the activation function in the fully connected layer, and a battery health state estimation model is generated. The pulse voltage response feature data is sequentially processed through the input embedding layer, multi-head self-attention layer, feed-forward network, and layer normalization layer of the Transformer regression model, and then global average pooling and mapping to the battery health state estimation range with a preset accuracy are performed to support precise adjustment of SOC / SOH conditions and adapt to random retirement scenarios, thereby enabling the generation of a battery health state estimation model.

[0166] Exemplarily, to facilitate understanding of the implementation process of the battery health state estimation method obtained by combining the above-mentioned Embodiment 1, please refer to Figure 7 , Figure 7 A schematic diagram of the overall process of a battery health state estimation method is provided. Specifically:

[0167] In the data collection and preprocessing stage, first, retired power battery samples are collected, 7-dimensional voltage features are extracted through pulse testing, and then a conditional vector database is constructed based on the extracted voltage features. The constructed conditional vector data is used as sample data, 70% of which is used for model training, and 30% is used for model verification.

[0168] Then, a physical constraint diffusion model is constructed, including a forward diffusion layer, a reverse denoising network, and a loss function jointly constructed by physical constraints and diffusion.

[0169] Finally, a battery health state estimation model based on the Transformer regression model is constructed, specifically including a data preprocessing layer, an input embedding layer, a multi-head self-attention layer, a feed-forward network, residual connections and layer normalization, global average pooling, and a fully connected layer. Thus, the battery health state estimation model is constructed.

[0170] It should be noted that in combination with Figure 8Describe the hardware calculation process of the SOH fast and low-power consumption estimation model. A computational graph is generated by the neural network model. In the computational graph, unblocked operator nodes are searched for. A computational model ability model is generated based on the hardware characteristics. The two are combined to calculate the block length of the input tensor, and the operators are blocked onto heterogeneous devices. At this time, it is judged whether there are unblocked nodes. If there are, continue to search for unblocked operator nodes; if not, the process ends and the prediction result is output.

[0171] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the battery health state estimation method of this application. Based on this technical concept, more forms of simple transformations are within the protection scope of this application.

[0172] This application also provides a battery health state estimation device. Please refer to Figure 9 , the battery health state estimation device includes:

[0173] A sample preprocessing module 10, configured to collect retired lithium-ion battery samples and preprocess the retired lithium-ion battery samples to obtain pulse voltage response data corresponding to different battery charge states;

[0174] A physical constraint diffusion module 20, configured to input the pulse voltage response data into a physical constraint diffusion model to generate pulse voltage response feature data;

[0175] An estimation model generation module 30, configured to generate a battery health state estimation model according to the pulse voltage response feature data and a Transformer regression model;

[0176] A health state estimation module 40, configured to estimate a battery to be evaluated according to the battery health state estimation model to obtain the battery health state corresponding to the battery to be evaluated.

[0177] The battery health state estimation device provided by this application adopts the battery health state estimation method in the above embodiment and can solve technical problems. Compared with the prior art, the beneficial effects of the battery health state estimation device provided by this application are the same as those of the battery health state estimation method provided by the above embodiment, and other technical features in the battery health state estimation device are the same as the features disclosed in the method of the above embodiment, which will not be elaborated here.

[0178] This application provides a battery health state estimation device. The battery health state estimation device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the battery health state estimation method in the first embodiment above.

[0179] Refer to the following Figure 10 , which shows a schematic structural diagram of a battery state of health estimation device suitable for implementing the embodiments of the present application. The battery state of health estimation device in the embodiments of the present application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistant), PADs (Portable Application Description: tablet computers), PMPs (Portable Media Player), in-vehicle terminals (such as in-vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 10 The shown battery state of health estimation device is merely an example and should not impose any limitation on the functions and usage scope of the embodiments of the present application.

[0180] As Figure 10 shown, the battery state of health estimation device may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. In the random access memory 1004, various programs and data required for the operation of the battery state of health estimation device are also stored. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are connected to each other through a bus 1005. An input / output interface 1006 is also connected to the bus. Generally, the following systems may be connected to the input / output interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the battery state of health estimation device to communicate with other devices wirelessly or wiredly to exchange data. Although the figure shows a battery state of health estimation device having various systems, it should be understood that it is not required to implement or have all the shown systems. More or fewer systems may be alternatively implemented or had.

[0181] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product that includes a computer program carried on a computer-readable medium, and the computer program contains program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by a processing device 1001, the above-mentioned functions defined in the methods of the embodiments disclosed in the present application are executed.

[0182] The battery health state estimation device provided by the present application adopts the battery health state estimation method in the above embodiments, and can solve the technical problems of battery health state estimation. Compared with the prior art, the beneficial effects of the battery health state estimation device provided by the present application are the same as those of the battery health state estimation method provided by the above embodiments, and other technical features in the battery health state estimation device are the same as those disclosed in the method of the previous embodiment, and will not be elaborated here.

[0183] It should be understood that the various parts disclosed in the present application can be implemented by hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0184] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0185] The present application provides a computer-readable storage medium having computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the battery health state estimation method in the above embodiments.

[0186] The computer-readable storage medium provided by this application can be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or components, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this embodiment, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, device, or component. The program code contained on the computer-readable storage medium can be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0187] The above computer-readable storage medium can be included in the battery health state estimation device; or it can exist independently without being assembled into the battery health state estimation device.

[0188] The above computer-readable storage medium carries one or more programs. When the above one or more programs are executed by the battery health state estimation device, the battery health state estimation device is caused to: collect retired lithium-ion battery samples, preprocess the retired lithium-ion battery samples to obtain pulse voltage response data corresponding to different battery charge states; input the pulse voltage response data into a physical constraint diffusion model to generate pulse voltage response feature data; generate a battery health state estimation model based on the pulse voltage response feature data and a Transformer regression model; and estimate the battery to be evaluated according to the battery health state estimation model to obtain the battery health state corresponding to the battery to be evaluated.

[0189] Computer program code for performing the operations of this application can be written in one or more programming languages or combinations thereof. The above-mentioned programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0190] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in the flowchart or block diagram can represent a module, a program segment, or a part of the code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks can occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks shown can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0191] The modules described in the embodiments of this application can be implemented in software or in hardware. Among them, the name of the module does not constitute a limitation on the unit itself in some cases.

[0192] The readable storage medium provided by this application is a computer-readable storage medium. The computer-readable storage medium stores computer-readable program instructions (i.e., computer programs) for performing the above-mentioned battery health state estimation method and can solve technical problems. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by this application are the same as those of the battery health state estimation method provided by the above embodiments and will not be elaborated here.

[0193] The present application also provides a computer program product, including a computer program which, when executed by a processor, implements the steps of the battery health state estimation method as described above.

[0194] The computer program product provided by the present application can solve technical problems. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as those of the battery health state estimation method provided by the above embodiments, and will not be elaborated herein.

[0195] The above are only partial embodiments of the present application, and thus do not limit the patent scope of the present application. Any equivalent structural transformation made under the technical concept of the present application by using the content of the specification and drawings of the present application, or any direct / indirect application in other related technical fields, is included in the patent protection scope of the present application.

Claims

1. A method for estimating a battery health state, characterized in that: The method includes: Collect retired lithium-ion battery samples and pre-process the retired lithium-ion battery samples to obtain pulse voltage response data corresponding to different battery charging states; Inputting the pulse voltage response data into a physical constrained diffusion model to generate pulse voltage response characteristic data; Generate a battery health state estimation model based on the pulse voltage response characteristic data and the Transformer regression model; The battery to be evaluated is estimated according to the battery health state estimation model to obtain the battery health state corresponding to the battery to be evaluated.

2. The method according to claim 1, characterized in that The step of inputting the pulse voltage response data into a physical constrained diffusion model to generate pulse voltage response characteristic data comprises: Determining noise data of the pulse voltage response data by forward diffusion; Inputting the noise data into a physical constraint diffusion model to perform physical constraint denoising to obtain theoretical voltage response data; determining a loss function according to the noise data and the theoretical voltage response data; Training the physical constrained diffusion model according to the loss function to obtain a trained model; Predicting the theoretical voltage response data according to the trained model to obtain predicted noise; The pulse voltage response data is subjected to reverse denoising according to the predicted noise to generate pulse voltage response characteristic data.

3. The method according to claim 2, characterized in that The physical constraint diffusion model includes an input layer, a backbone network and a physical constraint prediction head; The step of inputting the noise data into a physical constraint diffusion model to perform physical constraint denoising to obtain theoretical voltage response data comprises: Inputting the noise data into a physical constrained diffusion model, and determining a time step embedding and a conditional embedding corresponding to the noise data through the input layer; Performing sampling operations on the time step embedding and the conditional embedding through the backbone network to obtain sampled data; Predicting equivalent circuit model parameters corresponding to the sampled data by the physical constraint prediction head, and determining physical consistency loss based on the equivalent circuit model parameters and the pulse voltage response data; Physical constraint denoising is performed on the noise data according to the physical consistency loss to obtain theoretical voltage response data.

4. The method according to claim 3, characterized in that The backbone network includes a downsampling layer, an intermediate layer and an upsampling layer, the downsampling layer and the upsampling layer are connected through the intermediate layer, and the downsampling layer includes a residual block and a cross attention block; The step of performing a sampling operation on the time step embedding and the conditional embedding through the backbone network to obtain sampled data comprises: splicing the input of the residual block, the time step embedding and the conditional embedding through the cross attention block in the downsampling layer to obtain the spliced ​​input of the residual block; Inputting the concatenated residual block to the residual block in the downsampling layer to perform a maximum pooling operation to compress the feature map size, thereby obtaining the downsampled features; Capturing the global temporal dependency in the downsampled features through the intermediate layer to obtain intermediate features; The upsampling layer performs a feature map size restoration operation on the intermediate features to obtain sampled data.

5. The method according to claim 1, characterized in that The step of generating a battery health state estimation model according to the pulse voltage response characteristic data and the Transformer regression model comprises: Inputting the pulse voltage response characteristic data into a Transformer regression model to generate encoded characteristic data; Performing a global average pooling operation on the encoded feature data to obtain pooled feature data; The pooled feature data is mapped to a battery health state estimation range of a preset accuracy through an activation function in a fully connected layer to generate a battery health state estimation model.

6. The method according to claim 5, characterized in that The Transformer regression model includes an input embedding layer, a multi-head self-attention layer, a feedforward network, and a layer normalization layer; The step of inputting the pulse voltage response characteristic data into the Transformer regression model to generate encoded characteristic data comprises: Performing a standardization operation on the pulse voltage response characteristic data, and performing a normalization operation on the standardized data to obtain preprocessed characteristic data; Mapping the preprocessed feature data to a high-dimensional space through the input embedding layer to obtain high-dimensional feature data; Performing concatenation and linear transformation operations on the high-dimensional feature data through the multi-head self-attention layer to obtain transformed feature data; Performing nonlinear activation on the transformed feature data through the feedforward network to obtain feature data after nonlinear activation; The transformed feature data and the feature data after nonlinear activation are subjected to a residual connection operation through the layer normalization layer, and the connected feature data are subjected to a layer normalization operation to obtain encoded feature data.

7. A battery health status estimation device, characterized in that: The battery health status estimation device comprises: A sample preprocessing module is used to collect retired lithium-ion battery samples and preprocess the retired lithium-ion battery samples to obtain pulse voltage response data corresponding to different battery charging states; A physical constrained diffusion module, used for inputting the pulse voltage response data into a physical constrained diffusion model to generate pulse voltage response characteristic data; An estimation model generation module, used to generate a battery health state estimation model based on the pulse voltage response characteristic data and the Transformer regression model; The health state estimation module is used to estimate the battery to be evaluated according to the battery health state estimation model to obtain the battery health state corresponding to the battery to be evaluated.

8. A battery health status estimation device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the battery health state estimation method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the battery health state estimation method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the battery health state estimation method according to any one of claims 1 to 6 are implemented.

Images