Lithium battery health state estimation method and system based on confrontation and defense strategy

Abstract

The invention discloses a lithium battery health state estimation method and system based on a defense strategy, and the method comprises the steps: obtaining the voltage and current of a to-be-detected lithium battery, carrying out the ampere-hour integration of the current of a part of segments, and obtaining a curve of the voltage and capacity of the part of segments; based on the curves of the voltages and the capacities of the partial segments, selecting the capacities corresponding to a plurality of voltages on the curves; the capacities corresponding to the voltage values of the selected curve partial segments are input to a preset lithium battery health state estimation model, a lithium battery health state estimation value is obtained, and the preset lithium battery health state estimation model is obtained through training based on a convolutional residual network and an attack defense strategy. The system comprises a data acquisition module, a data selection module and a data processing module. According to the method, the SOH of a normal input sample can be more accurately estimated, and meanwhile, the method still has more accurate SOH estimation for an input sample against resistance attacks.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium battery health state estimation, and in particular relates to a method and system for lithium battery health state estimation based on an adversarial defense strategy. Background Art

[0002] Lithium-ion batteries have become an integral part of the electrochemical energy storage landscape and are widely used in electric vehicles and battery energy storage systems. However, the challenge of battery aging remains unavoidable, rooted in its complex and multifaceted degradation mechanisms, including lithium inventory loss, active material loss, and conductivity loss. In order to ensure the safe operation and power capabilities of lithium-ion batteries, accurate monitoring of the state of health (SOH) of lithium batteries is required.

[0003] As a "black box" model, the deep learning method can extract and analyze potential rules from the recorded historical data sets. In this way, there is no need to have a deep understanding of the potential electrochemical reactions inside the lithium battery, which has obvious advantages in terms of applicability.

[0004] The three existing solutions that are closest to this patent application are as follows:

[0005] CN117741484A A method for estimating the health status of a lithium battery based on the characteristics of a charging voltage curve. The invention discloses a method for estimating the health status of a lithium battery based on the characteristics of a charging voltage curve, and relates to the technical field of lithium battery state prediction. It includes: S1 obtaining a charging voltage data sequence and a health status data sequence of each charging and discharging cycle of a lithium battery; S2 determining a constant current charging time period, and calculating a difference ΔV-i between a charging voltage curve and an average value of a charging voltage in the constant current charging time period; S3 using the Pearson correlation coefficient to analyze the correlation between the ΔV-i obtained in S2 and the current maximum charging capacity of the battery; S4 collecting training samples to train a convolutional neural network-long short-term memory network to obtain a lithium battery SOH estimation model; S5 using the lithium battery SOH estimation model to estimate the SOH of a lithium battery. The invention can propose a feature that effectively characterizes the degradation of a lithium battery from a voltage curve of a relatively fixed charging process of a lithium battery, thereby improving the accuracy of a lithium battery SOH prediction model.

[0006] CN116774086B A lithium battery health state estimation method based on multi-sensor data fusion. The invention discloses a lithium battery health state estimation method based on multi-sensor data fusion, collects lithium battery data, uses a detrended cross-correlation analysis method to extract features from a data set and divides the data set; constructs a lithium battery pack battery health state model based on STGCN and Pyraformer models; extracts local battery capacity change features of a spatial parameter sequence composed of multi-dimensional parameter data of a lithium battery pack through STGCN, and inputs the obtained features into Pyraformer to establish a change relationship between the spatial parameter sequence and the global lithium battery pack health state; uses uniform initialization to initialize SDO, introduces improved multivariate learning into SDO to obtain an improved UMSDO algorithm; uses UMSDO to optimize hyperparameters in the STGCN-Pyraformer model, obtains corresponding optimal parameters, predicts the battery health state, and obtains a prediction result of the battery health state. The invention can be applied to the modeling process of lithium battery health state prediction to ensure the accuracy of lithium battery health state prediction.

[0007] CN116908692A A strong and robust method for jointly estimating the state of charge and health of lithium batteries. The present invention relates to a strong and robust method for jointly estimating the state of charge and health of lithium batteries. The minimum error entropy is used to replace the minimum mean square error in the traditional algorithm. Secondly, the maximum correlation entropy is combined with MEE to design the reference point to obtain MEEF. MEEF can adjust the error average value to zero, further effectively handle non-Gaussian noise in complex environments, and finally derive the minimum error entropy bidirectional square root volumetric Kalman filter (MEEF-DSRCKF) algorithm based on the reference point. This patent takes the second-order equivalent circuit model as an example, uses internal resistance to measure SOH, and realizes the joint estimation of lithium batteries by updating two MEEF-SRCKF filters with each other. Finally, under non-Gaussian noise, the MEEF-DSRCKF algorithm, the MEE-DSRCKF algorithm and the traditional DSRCKF algorithm are compared. The results show that the present invention further improves the high accuracy and strong robustness of the estimation of the state of charge and health of the battery under non-Gaussian noise.

[0008] The above inventions 1 to 3 are all battery SOH assessments, and all three provide methods for assessing battery SOH. Through different technical means and algorithms, the battery SOH is estimated based on different battery characteristics such as: charging voltage curve characteristics, partial battery capacity change characteristics, and internal resistance characteristics. The invention aims to improve the accuracy and practicality of battery SOH assessment; different data processing, model construction and algorithm frameworks are used to more accurately estimate the battery SOH. Invention 1 calculates the difference ΔV-i between the charging voltage curve and the average value of the charging voltage during the constant current charging time period as a feature, and invention 2 extracts the local battery capacity change characteristics of the spatial parameter sequence composed of multi-dimensional parameter data of the lithium battery pack. These two inventions can achieve accurate estimation of battery SOH, but lack consideration of model robustness. Invention 3 takes the second-order equivalent circuit model as an example and uses internal resistance to measure SOH. Although this method can handle non-Gaussian noise in complex environments, it has poor practicality and generalization. The purpose of the invention is to provide technical support for the reliable operation of energy storage batteries. By accurately evaluating the SOH of the battery, the battery usage strategy can be optimized, the battery life can be extended, the risk of failure can be reduced, and a basis can be provided for the design and management of the energy storage system. The invention can be used for the management and operation of the energy storage system. Through accurate evaluation and online monitoring of the SOH of the energy storage battery, the system manager can optimize the scheduling, capacity planning and fault diagnosis according to the actual condition of the battery to ensure the stable operation of the energy storage system. By evaluating and monitoring the SOH of the energy storage battery, the battery usage and maintenance strategy can be optimized, and the overcharging and discharging and adverse operation of the battery can be reduced, thereby extending the battery life and cycle number. These inventions can be used to detect battery failures and abnormal conditions early. Through online monitoring of the battery SOH, potential fault characteristics and abnormal behaviors can be identified, and early warning information can be provided so that appropriate maintenance and replacement measures can be taken to avoid the impact of battery failures on the energy storage system.

[0009] However, when the above invention method is actually deployed, the input sample data will not only be affected by the sensor measurement, but also may be tampered by some unscrupulous attackers after being uploaded to the cloud battery management system (BMS), resulting in incorrect SOH estimation results. Adversarial attacks bring new attack opportunities to attackers. Adversarial attacks slightly perturb the input sample data to change the results of SOH estimation. Deep learning models are easily affected by adversarial attacks, and the robustness of deep learning models will be greatly reduced under adversarial attacks. Once a cyber-physical attack occurs, the battery state may be modified to an incorrect value, which may increase the safety risk of energy storage systems with thousands of batteries. Attackers can modify the original input and guide the trained deep learning model to an incorrect state, resulting in dangerous health management decisions. For example, if the attacker sets the battery SOH to 0.9, and the actual SOH has reached 0.8, the user will think that it is safe to use lithium batteries. Further triggering thermal runaway or even serious safety accidents. If the SOH estimation model using deep learning does not have sufficient safety measures to deal with adversarial attacks, it is vulnerable to malicious attacks, resulting in SOH estimation errors, exacerbating the complexity of maintaining the health and safety of battery energy storage systems or electric vehicles. Therefore, adversarial attacks pose new challenges to deep learning models, requiring them to have not only the accuracy of SOH estimation but also the robustness to adversarial attacks. Summary of the invention

[0010] The purpose of the present invention is to provide a method and system for estimating the health status of a lithium battery based on an adversarial defense strategy, which can accurately estimate the SOH for normal input samples, and at the same time still have a relatively accurate SOH estimation for input samples subjected to adversarial attacks, so that the model has both the accuracy of SOH estimation and the robustness against attacks, thereby preventing the model from being manipulated by a small number of criminals.

[0011] In order to achieve the above object, the present invention adopts the following technical solution:

[0012] The method for estimating the health status of lithium batteries based on the adversarial defense strategy includes:

[0013] The voltage and current of the lithium battery to be tested are obtained, and the current of a part of the fragment is integrated by ampere-hour to obtain a curve of the voltage and capacity of the part of the fragment;

[0014] Based on the voltage and capacity curve of some segments, select the capacities corresponding to several voltages on the curve;

[0015] The capacities corresponding to several voltage values ​​of the selected curve segment are input into a preset lithium battery health state estimation model to obtain a lithium battery health state estimation value, wherein the preset lithium battery health state estimation model is trained based on a convolutional residual network and a defense strategy against attacks.

[0016] A further improvement of the present invention is to obtain the voltage and current of the lithium battery to be tested, perform ampere-hour integration on the current of a partial segment, and obtain a curve of the voltage and capacity of the partial segment, including:

[0017] Among them, for some fragments V c The calculation formula used for the ampere-hour integration of the current is:

[0018]

[0019] Where Q(V ci ) is V c The i-th voltage V selected on ci The corresponding capacity, t Vci , t Vstart The voltage V ci and V c Starting voltage V start The corresponding time, I(t) is the current.

[0020] A further improvement of the present invention is that, based on the curve of voltage and capacity of some segments, the capacities corresponding to several voltages on the curve are selected, including:

[0021] Part of the fragment V c Divided into N = (V end -V start ) / ΔV+1 segments, ΔV represents the voltage interval of each segment, V end V c The end voltage of the voltage sequence [V c1 ,V c2 ,V c3 ,...,V cN ]=[V start ,V start +ΔV,...,V start +(N-1)×ΔV], and the capacity Q(V c )=[Q(V c1 ),Q(V c2 ),...,Q(V cN )].

[0022] A further improvement of the present invention is that the preset lithium battery health state estimation model is obtained by training based on a convolutional residual network and a defense strategy against attacks, wherein the convolutional residual network includes M convolutional blocks, a global average pooling layer and a fully connected layer, the input forms a feature map through the convolutional block, and the global average pooling layer is used to reduce the dimension of the feature map. Finally, the fully connected layer is used to map the reduced feature map to the battery SOH.

[0023] A further improvement of the present invention is that the preset lithium battery health state estimation model is obtained by training based on a convolutional residual network and a defense strategy against attacks, wherein the defense strategy against attacks includes:

[0024] In each round of convolutional residual network training, the generated adversarial samples are added to the training dataset so that the model can learn the relationship between the adversarial samples and their corresponding outputs. The trained model adapts to the changes from the original input to the adversarial samples. The optimization goal is expressed as follows:

[0025]

[0026] The inner maximization stage seeks to maximize the original input x given the model. 0 The external minimization problem aims to determine the parameter θ that minimizes the loss of the internal normal samples and the found adversarial samples. * , thereby enhancing the overall robustness of the model; L is the loss function, y is the true SOH, f θ is a convolutional residual network with parameter θ, and δ is the perturbation radius.

[0027] The system for estimating the health status of lithium batteries based on the adversarial defense strategy includes:

[0028] The data acquisition module acquires the voltage and current of the lithium battery to be tested, integrates the current of a part of the fragment in ampere-hours, and obtains the curve of the voltage and capacity of the part of the fragment;

[0029] A data selection module selects the capacities corresponding to several voltages on the curve based on the voltage and capacity curve of some segments;

[0030] The data processing module inputs the capacity corresponding to several voltage values ​​of the selected curve segment into a preset lithium battery health state estimation model to obtain a lithium battery health state estimation value, wherein the preset lithium battery health state estimation model is obtained by training based on a convolutional residual network and a defense strategy against attacks.

[0031] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for estimating the health status of a lithium battery based on an adversarial defense strategy.

[0032] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0033] The method and system for estimating the health status of lithium batteries based on the adversarial defense strategy provided by the present invention can achieve accurate SOH estimation for normal samples using a neural network with an adversarial training strategy. A neural network using a conventional training strategy can only achieve accurate SOH estimation for normal samples, but when encountering adversarial samples, the SOH estimation result is poor, while a neural network using an adversarial training strategy can also achieve relatively accurate SOH estimation for adversarial samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the structure of RCN;

[0035] Figure 2 Schematic diagram of SOH estimation results for normal samples and adversarial samples;

[0036] Figure 3 Comparison diagram of normal samples and adversarial samples in the initial and final cycles of NCM#1;

[0037] Figure 4 Schematic diagram of SOH estimation results for adversarial examples during normal training and adversarial training;

[0038] Figure 5 The structure block diagram of the system for estimating the health status of lithium batteries based on the countermeasure defense strategy of the present invention. DETAILED DESCRIPTION

[0039] In the following, only some exemplary embodiments are briefly described. As those skilled in the art will appreciate, the described embodiments may be modified in various ways without departing from the spirit or scope of the present invention. Therefore, the drawings and descriptions are considered to be exemplary and non-restrictive in nature.

[0040] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0041] It should also be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0042] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0043] Various structural schematic diagrams of the embodiments disclosed in the present invention are shown in the accompanying drawings. These figures are not drawn to scale, and some details are magnified and some details may be omitted for the purpose of clear expression. The shapes of various regions and layers shown in the figures and the relative sizes and positional relationships therebetween are only exemplary, and may deviate in practice due to manufacturing tolerances or technical limitations, and those skilled in the art may additionally design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0044] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0045] Example 1

[0046] The method for estimating the health status of a lithium battery based on an adversarial defense strategy provided by the present invention includes:

[0047] The voltage and current of the lithium battery to be tested are obtained, and the current of a part of the fragment is integrated by ampere-hour to obtain a curve of the voltage and capacity of the part of the fragment;

[0048] Based on the voltage and capacity curve of some segments, select the capacities corresponding to several voltages on the curve;

[0049] The capacities corresponding to several voltage values ​​of the selected curve segment are input into a preset lithium battery health state estimation model to obtain a lithium battery health state estimation value, wherein the preset lithium battery health state estimation model is trained based on a convolutional residual network and a defense strategy against attacks.

[0050] In this embodiment, the preset lithium battery health state estimation model is obtained by training based on a convolutional residual network and a defense strategy against adversarial attacks, wherein the convolutional residual network includes M convolutional blocks, a global average pooling layer and a fully connected layer, the input forms a feature map through the convolutional block, and the global average pooling layer is used to reduce the dimension of the feature map. Finally, the fully connected layer is used to map the reduced feature map to the battery SOH.

[0051] Defense strategies to counter attacks include:

[0052] In each round of convolutional residual network training, the generated adversarial samples are added to the training dataset so that the model can learn the relationship between the adversarial samples and their corresponding outputs. The trained model adapts to the changes from the original input to the adversarial samples. The optimization goal is expressed as follows:

[0053]

[0054] The inner maximization stage seeks to maximize the original input x given the model. 0The external minimization problem aims to determine the parameter θ that minimizes the loss of the internal normal samples and the found adversarial samples. * , thereby enhancing the overall robustness of the model; L is the loss function, y is the true SOH, f θ is a convolutional residual network with parameter θ, and δ is the perturbation radius.

[0055] Example 2

[0056] The present invention provides a method for estimating the health status of lithium batteries based on an adversarial defense strategy. The present invention uses the coulomb counting of a partial voltage segment as a health feature. The proposed convolutional residual network (RCN) model is used to estimate the battery SOH. Figure 1 The basic elements of the RCN model in include M convolutional blocks (Conv Blocks), global average pooling (GAP) layers, and fully connected (FC) layers. The input is formed into a feature map through Conv Blocks, and the Conv Blocks with residual connections can better extract the features behind the input. Then, the GAP layer is used to reduce the dimension of the feature map. Finally, the FC layer is used to map the reduced feature map to the battery SOH.

[0057] Conv Blocks contain residual connections that preserve the initial input within each block, thus avoiding the vanishing or exploding gradients in traditional neural networks and ensuring the accuracy of battery SOH estimation. Figure 1 The Conv block in contains three key components: Convolutional (Conv) layer, Batch Normalization (BN) layer, and Rectified Linear Unit (ReLU) layer. The Conv layer is responsible for extracting features from the input data through the convolution kernel. Considering the time series nature of the input data, a one-dimensional convolution method is selected at this stage. The BN layer plays a key role in maintaining a stable input distribution for each subsequent layer, which helps to alleviate the gradient vanishing and speed up the network convergence. ReLU acts as a nonlinear activation function.

[0058] Adversarial attacks refer to the situation where the attacker introduces subtle perturbations to the original data to generate adversarial samples that may affect the results of the deep learning model. When exposed to the designed adversarial samples, these attacks may cause the trained deep learning model to produce a large number of erroneous results. This manipulation can have a significant impact on the battery SOH estimation. Projected Gradient Descent (PGD) was selected as the adversarial attack method, which is a gradient-based attack and uses the gradient of the loss function to iteratively introduce perturbations to the input. PGD is considered to be one of the most powerful first-order adversarial attack techniques and poses a major challenge to the robustness of deep learning models. By successfully defending against PGD attacks, the model can address other first-order adversarial attack methods, thereby enhancing the robustness of the model. The goal of the deep learning model is to minimize the loss function L using the gradient descent method, while the goal of the PGD adversarial attack is to manipulate the input so that the SOH estimation result deviates as much as possible from the actual SOH, so the input of the PGD attack is to rise in the opposite direction of the gradient descent. The input of the PGD attack is calculated by the following formula:

[0059]

[0060] Where x represents the input, ε represents the step size, and x t represents the adversarial sample input after the tth iteration, x 0 represents the original input, i.e., the normal sample, Δx represents the acceptable perturbation radius, and the adversarial sample x t Projection back to x 0 +Δxl ∞ Spherical surface.

[0061] Adversarial training is the main strategy for defending against adversarial attacks and can enhance the robustness of deep learning models. For a deep learning framework with regular training, the core goal is to determine the optimal model parameters θ that minimize the loss function over the entire training dataset. * (Equation (3)). This process aims to achieve a model with both high accuracy and good generalization ability when facing normal samples.

[0062]

[0063] Conventional normal training has demonstrated its effectiveness in various tasks, but often lacks robustness in the face of adversarial examples. This limitation leads to the need to improve conventional normal training methods.

[0064] In each round of training, by adding the generated adversarial examples to the training dataset, the model can learn the relationship between the adversarial examples and their corresponding outputs. In this way, the trained model can adapt to subtle changes from the original input to the adversarial examples, thereby enhancing the robustness of the model. The optimization objective can be restated as follows:

[0065]

[0066] The inner maximization stage seeks to maximize the original input x given the model. 0 The outer minimization problem then aims to determine the parameter θ that minimizes the loss of the inner normal samples and the found adversarial examples. * , thereby enhancing the overall robustness of the model; y is the true SOH, f θ is a convolutional residual network with parameter θ, and δ is the perturbation radius.

[0067] Example 3

[0068] The present invention is verified on NCM batteries, and 66 NCM batteries are divided into a training set and a test set in a ratio of 4:1, wherein the test set contains 14 NCM batteries numbered NCM#1-NCM#14.

[0069] First, extract some segment features of the battery. Specifically, the charging voltage segment V c For SOH estimation, V c The start and end of start and V end . V c Divided into N = (V end -V start ) / ΔV+1 segments, ΔV represents the voltage interval of each segment. Therefore, we get the sequence [V c1 ,V c2 ,V c3 ,...,V cN ]=[V start ,V start +ΔV,...,V start +(N-1)×ΔV].

[0070] Then, it is expressed as Q(V c )=[Q(V c1 ),Q(V c2 ),...,Q(V cN )], and Q(V ci ) is determined by equation (1).

[0071]

[0072] Among them, t Vci , t Vstart V ci and V start The corresponding time, I(t) is the current, it should be noted that Q(V c ) can be extracted from the partial charge cycle and can be directly obtained from the BMS.c The range is V start =3.55V to V end =3.74V, ΔV=0.01V, N=20.

[0073] Then the proposed RCN model is trained using the conventional normal training strategy, and PGD is used to generate the corresponding adversarial attack samples. The hyperparameter configuration of RCN is shown in the following table:

[0074] Table 1 Hyperparameter configuration of RCN

[0075]

[0076] The parameters of PGD are shown in the following table:

[0077] Table 2 PGD hyperparameter configuration

[0078] Hyperparameters value Disturbance radius Δx 0.01 Iteration steps 50 ε 0.0004 a 0.8 b 1 β 1000

[0079] The hyperparameters for normal training are shown in the following table:

[0080] Table 3 Hyperparameters for normal training

[0081]

[0082]

[0083] Taking NCM#1 in the test set as an example, the SOH estimation results of the RCN model after conventional normal training for normal samples and adversarial samples are as follows: Figure 2 As shown:

[0084] It can be seen from the figure that in the absence of adversarial attacks, the estimated SOH is very accurate and very close to the reference value, which proves the effectiveness of the proposed RCN. However, when the system is subjected to adversarial attacks, the SOH estimation deviates significantly from the reference SOH. In addition, Figure 3 (a) and (b) show the comparison between normal input and PGD adversarial attack input at the initial life cycle and the final life cycle of NCM#1. Figure 3 As shown in Figure 2, adversarial samples and normal samples obtained from PGD attacks are very close, which means it is difficult to visually distinguish whether the input is attacked. However, the SOH estimation results of adversarial samples and normal samples show obvious differences, highlighting the far-reaching impact of PGD attacks.

[0085] Meanwhile, Table 4 details the average numerical data for all 14 test set batteries. The MAE and RMSE of SOH estimation for normal samples are very low, 0.418% and 0.654%, respectively. These values ​​indicate that the accuracy of SOH estimation is very high when not subjected to adversarial attacks. In contrast, adversarial samples generated by PGD exhibit larger errors in SOH estimation. The MAE and RMSE produced by adversarial samples generated by PGD are 12.6% and 14.3%, respectively. These error metrics are 30.14 times and 21.87 times the error metrics observed in normal samples, respectively, indicating a significant drop in SOH estimation accuracy. These huge differences demonstrate the adverse effects of adversarial attacks on the accuracy of SOH estimation.

[0086] Table 4 SOH estimation results of adversarial samples and normal samples for conventional normal training

[0087] Normal sample Adversarial Examples MAE(%) 0.418 12.6 RMSE(%) 0.654 14.3

[0088] The above results emphasize that adversarial examples can cause SOH estimation errors, so adversarial training is needed as a defense strategy to improve the original model. In addition, adversarial examples generated using different parameter settings have a significant impact on battery SOH estimation. Here we study two parameters in depth: perturbation radius Δx and number of iterations.

[0089] right Figure 4 Comparative analysis of the RMSE and MAE of adversarial examples with the same parameter set in (a) / (c) and (b) / (d) reveals the effectiveness of the adversarial training defense strategy in reducing the SOH estimation error caused by adversarial examples. Even in scenarios where the adversarial example produces a maximum RMSE = 0.2445 for the conventional normal training model, such as 0.019 perturbation radius and 10 iteration steps, the SOH estimation error of the adversarially trained RCN model is significantly reduced to 0.037. This is approximately equivalent to 15% of the RMSE under conventional normal training conditions, which significantly demonstrates the effectiveness of adversarial training defense in enhancing model robustness. And for normal samples, the results of the SOH estimation RMSE and MAE of the adversarially trained RCN model are 0.01890 and 0.01176, respectively, indicating that the model can achieve accurate SOH estimation for both normal and adversarial samples.

[0090] Example 4

[0091] like Figure 5 As shown, the system for estimating the health status of a lithium battery based on an adversarial defense strategy provided by the present invention includes:

[0092] The data acquisition module acquires the voltage and current of the lithium battery to be tested, integrates the current of a part of the fragment in ampere-hours, and obtains the curve of the voltage and capacity of the part of the fragment;

[0093] A data selection module selects the capacities corresponding to several voltages on the curve based on the voltage and capacity curve of some segments;

[0094] The data processing module inputs the capacity corresponding to several voltage values ​​of the selected curve segment into a preset lithium battery health state estimation model to obtain a lithium battery health state estimation value, wherein the preset lithium battery health state estimation model is obtained by training based on a convolutional residual network and a defense strategy against attacks.

[0095] Example 5

[0096] The present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the method for estimating the health status of a lithium battery based on an adversarial defense strategy are implemented.

[0097] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0098] The present application is described with reference to the flowcharts and / or block diagrams of the methods, systems and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A system that specifies the functions of a box or multiple boxes.

[0099] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0100] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0101] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the attached claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved.

[0102] In addition, it should be understood that although this specification is described in accordance with the implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation modes that can be understood by those skilled in the art. The above content is only to illustrate the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution according to the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A method for estimating the health status of a lithium battery based on an adversarial defense strategy, characterized in that: include: The voltage and current of the lithium battery to be tested are obtained, and the current of a part of the fragment is integrated in ampere-hour to obtain a curve of the voltage and capacity of the part of the fragment; Based on the voltage and capacity curve of some segments, select the capacities corresponding to several voltages on the curve; The capacities corresponding to several voltage values ​​of the selected curve segment are input into a preset lithium battery health state estimation model to obtain a lithium battery health state estimation value, wherein the preset lithium battery health state estimation model is trained based on a convolutional residual network and a defense strategy against attacks.

2. The method for estimating the health status of a lithium battery based on an adversarial defense strategy according to claim 1, characterized in that: The voltage and current of the lithium battery to be tested are obtained, and the current of a part of the fragment is integrated by ampere-hour to obtain the curve of the voltage and capacity of the part of the fragment, including: For some fragments V c The calculation formula used for the ampere-hour integration of the current is: Where Q(V ci ) is V c The i-th voltage V selected on ci The corresponding capacity, t Vci , t Vstart The voltage V ci and V c Starting voltage V start The corresponding time, I(t) is the current.

3. The method for estimating the health status of a lithium battery based on an adversarial defense strategy according to claim 2, characterized in that: Based on the voltage and capacity curve of some segments, select the capacities corresponding to several voltages on the curve, including: Part of the fragment V c Divided into N = (V end -V start ) / ΔV+1 segments, ΔV represents the voltage interval of each segment, V end V c The end voltage of the voltage sequence [V c1 ,V c2 ,V c3 ,...,V cN ]=[V start ,V start +ΔV,...,V start +(N-1)×ΔV], and the capacity Q(V c )=[Q(V c1 ),Q(V c2 ),...,Q(V cN )].

4. The method for estimating the health status of a lithium battery based on an adversarial defense strategy according to claim 1, characterized in that: The preset lithium battery health state estimation model is obtained by training based on a convolutional residual network and a defense strategy against adversarial attacks. The convolutional residual network includes M convolutional blocks, a global average pooling layer and a fully connected layer. The input forms a feature map through the convolutional block, and a global average pooling layer is used to reduce the dimension of the feature map. Finally, a fully connected layer is used to map the reduced feature map to the battery SOH.

5. The method for estimating the health status of a lithium battery based on an adversarial defense strategy according to claim 4, characterized in that: The preset lithium battery health status estimation model is obtained by training based on a convolutional residual network and a defense strategy against adversarial attacks, wherein the defense strategy against adversarial attacks includes: In each round of convolutional residual network training, the generated adversarial samples are added to the training dataset so that the model can learn the relationship between the adversarial samples and their corresponding outputs. The trained model adapts to the changes from the original input to the adversarial samples. The optimization goal is expressed as follows: The inner maximization stage seeks to maximize the original input x given the model. 0 The external minimization problem aims to determine the parameter θ that minimizes the loss of the internal normal samples and the found adversarial samples. * , thereby enhancing the overall robustness of the model; L is the loss function, y is the true SOH, f θ is a convolutional residual network with parameter θ, and δ is the perturbation radius.

6. A system for estimating the health status of lithium batteries based on an adversarial defense strategy, characterized in that: include: The data acquisition module acquires the voltage and current of the lithium battery to be tested, integrates the current of a part of the fragment in ampere-hours, and obtains the curve of the voltage and capacity of the part of the fragment; A data selection module selects the capacities corresponding to several voltages on the curve based on the voltage and capacity curve of some segments; The data processing module inputs the capacity corresponding to several voltage values ​​of the selected curve segment into a preset lithium battery health state estimation model to obtain a lithium battery health state estimation value, wherein the preset lithium battery health state estimation model is obtained by training based on a convolutional residual network and a defense strategy against attacks.

7. The system for estimating the health status of a lithium battery based on an adversarial defense strategy according to claim 6, characterized in that: In the data acquisition module, the voltage and current of the lithium battery to be tested are obtained, and the current of some segments is integrated by ampere-hour to obtain the curve of voltage and capacity of some segments, including: For some fragments V c The calculation formula used for the ampere-hour integration of the current is: Where Q(V ci ) is V c The i-th voltage V selected on ci The corresponding capacity, t Vci , t Vstart The voltage V ci and V c Starting voltage V start The corresponding time, I(t) is the current.

8. The system for estimating the health status of a lithium battery based on an adversarial defense strategy according to claim 7, characterized in that: In the data selection module, based on the voltage and capacity curve of some segments, the capacities corresponding to several voltages on the curve are selected, including: Part of the fragment V c Divided into N = (V end -V start ) / ΔV+1 segments, ΔV represents the voltage interval of each segment, V end V c The end voltage of the voltage sequence [V c1 ,V c2 ,V c3 ,...,V cN ]=[V start ,V start +ΔV,...,V start +(N-1)×ΔV], and the capacity Q(V c )=[Q(V c1 ),Q(V c2 ),...,Q(V cN )].

9. The system for estimating the health status of a lithium battery based on an adversarial defense strategy according to claim 8, characterized in that: In the data processing module, the preset lithium battery health status estimation model is obtained by training based on the convolutional residual network and the defense strategy against adversarial attacks. The convolutional residual network includes M convolutional blocks, a global average pooling layer and a fully connected layer. The input forms a feature map through the convolutional block, and the global average pooling layer is used to reduce the dimension of the feature map. Finally, the fully connected layer is used to map the reduced feature map to the battery SOH.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for estimating the health status of a lithium battery based on an adversarial defense strategy according to any one of claims 1 to 5.

Images