Battery fault detection method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of battery fault detection, in particular to a battery fault detection method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring electrochemical impedance spectrum data of a single battery, and generating an electrochemical reaction spectrum according to a plurality of frequency points in the electrochemical impedance spectrum data and impedance real part data and impedance imaginary part data of each frequency point; calculating feature vectors of a plurality of reaction features in the electrochemical reaction spectrum, and performing clustering analysis on each feature vector to obtain clustering result data; screening the reaction features according to the clustering result data to obtain target features, and determining a feature weight value of each target feature; and determining a fault detection result of the single battery according to the feature weight value of each target feature and the feature vector of each target feature. By adopting the method, the battery fault detection accuracy in a small sample or boundary scene can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of battery fault detection, and in particular to a battery fault detection method, apparatus, computer equipment, and storage medium. Background Art

[0002] With the large-scale application of batteries in energy storage, electric transportation and other scenarios, their safety and health status monitoring has become one of the core issues in system operation and maintenance.

[0003] Electrochemical Impedance Spectroscopy (EIS) is a non-destructive testing method for battery status assessment. It can measure the battery's response to AC signals at multiple frequency points, thereby reflecting its internal reaction mechanism and health. In related technologies, EIS data modeling or combined with expert experience is used to identify battery status. However, such methods are highly dependent on physical modeling or manual experience, and therefore easily miss important features that are sensitive to early failures, lack adaptability and generalization capabilities, and are prone to misjudgment in small samples or boundary scenarios. Summary of the Invention

[0004] Based on this, it is necessary to provide a battery fault detection method, device, computer equipment, computer-readable storage medium and computer program product that can improve the accuracy of battery fault detection in small samples or boundary scenarios to address the above technical problems.

[0005] In a first aspect, the present application provides a battery fault detection method, comprising:

[0006] Acquire electrochemical impedance spectroscopy data of the battery cell, and generate an electrochemical reaction spectrum based on multiple frequency points in the electrochemical impedance spectroscopy data, as well as real impedance data and imaginary impedance data at each frequency point;

[0007] Calculating characteristic vectors of a plurality of reaction features in the electrochemical reaction spectrum, and performing cluster analysis on each characteristic vector to obtain clustering result data;

[0008] Screening the reaction features according to the clustering result data to obtain target features, and determining feature weight values ​​of each target feature;

[0009] A fault detection result of the battery cell is determined according to the feature weight value of each target feature and the feature vector of each target feature.

[0010] In one embodiment, screening the reaction features according to the clustering result data to obtain target features, and determining feature weight values ​​of each target feature, includes:

[0011] Determining the number of clusters for each of the feature vectors according to the clustering result data;

[0012] When the number of clusters is greater than or equal to a screening threshold, the corresponding reaction feature is used as a target feature;

[0013] Calculating the intra-class dispersion and inter-class dispersion of each target feature according to the clustering result data;

[0014] The feature weight value of each target feature is calculated according to the intra-class dispersion and the inter-class dispersion.

[0015] In one embodiment, determining the fault detection result of the battery cell according to the feature weight value of each target feature and the feature vector of each target feature includes:

[0016] Get historical battery data;

[0017] Calculating fault score data of the battery cell according to the feature vector of each target feature, the feature weight value and the historical battery data;

[0018] When the fault score data exceeds a fault threshold, it is determined that the battery cell is in a fault state.

[0019] In one embodiment, calculating the fault score data of the battery cell according to the feature vector of each target feature, the feature weight value and the historical battery data includes:

[0020] According to the feature weight values ​​of each of the target features, the feature vector is subjected to dimensionality reduction using a weighted principal component analysis method to obtain a first principal component vector and a second principal component vector; the feature weight values ​​corresponding to each of the target features in the first principal component vector are greater than the feature weight values ​​corresponding to each of the target features in the second principal component vector;

[0021] constructing a time series feature vector based on the first principal component vector and the historical battery data;

[0022] Fault score data of the battery cell is calculated according to the time series feature vector, the first principal component vector, and the second principal component vector.

[0023] In one embodiment, calculating the fault score data of the battery cell according to the time series feature vector, the first principal component vector, and the second principal component vector includes:

[0024] Inputting the time series feature vector, the first principal component vector, and the second principal component vector into a kernel density estimation model to obtain a probability density value; wherein the kernel density estimation model is constructed based on characteristic data of historical non-faulty battery cells;

[0025] Fault score data of the battery cell is calculated according to the probability density value.

[0026] In one embodiment, the method further comprises:

[0027] After the battery cell completes fault detection, adding the fault score data of the battery cell to an update set;

[0028] When the number of the battery cells included in the update set is greater than or equal to an update threshold, updating the fault threshold according to the fault score data of each battery cell in the update set;

[0029] Clear the update set.

[0030] In a second aspect, the present application further provides a battery fault detection device, comprising:

[0031] A data acquisition module, configured to acquire electrochemical impedance spectroscopy data of a battery cell and generate an electrochemical reaction spectrum based on multiple frequency points in the electrochemical impedance spectroscopy data and the real impedance data and imaginary impedance data of each frequency point;

[0032] A feature clustering module, configured to calculate feature vectors of a plurality of reaction features in the electrochemical reaction spectrum, and perform cluster analysis on each feature vector to obtain clustering result data;

[0033] A feature screening module is used to screen the reaction features according to the clustering result data to obtain target features and determine the feature weight value of each target feature;

[0034] A fault detection module is configured to determine a fault detection result of the battery cell according to the feature weight value of each target feature and the feature vector of each target feature.

[0035] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0036] Acquire electrochemical impedance spectroscopy data of the battery cell, and generate an electrochemical reaction spectrum based on multiple frequency points in the electrochemical impedance spectroscopy data, as well as real impedance data and imaginary impedance data at each frequency point;

[0037] Calculating characteristic vectors of a plurality of reaction features in the electrochemical reaction spectrum, and performing cluster analysis on each characteristic vector to obtain clustering result data;

[0038] Screening the reaction features according to the clustering result data to obtain target features, and determining feature weight values ​​of each target feature;

[0039] A fault detection result of the battery cell is determined according to the feature weight value of each target feature and the feature vector of each target feature.

[0040] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0041] Acquire electrochemical impedance spectroscopy data of the battery cell, and generate an electrochemical reaction spectrum based on multiple frequency points in the electrochemical impedance spectroscopy data, as well as real impedance data and imaginary impedance data at each frequency point;

[0042] Calculating characteristic vectors of a plurality of reaction features in the electrochemical reaction spectrum, and performing cluster analysis on each characteristic vector to obtain clustering result data;

[0043] Screening the reaction features according to the clustering result data to obtain target features, and determining feature weight values ​​of each target feature;

[0044] A fault detection result of the battery cell is determined according to the feature weight value of each target feature and the feature vector of each target feature.

[0045] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the following steps:

[0046] Acquire electrochemical impedance spectroscopy data of the battery cell, and generate an electrochemical reaction spectrum based on multiple frequency points in the electrochemical impedance spectroscopy data, as well as real impedance data and imaginary impedance data at each frequency point;

[0047] Calculating characteristic vectors of a plurality of reaction features in the electrochemical reaction spectrum, and performing cluster analysis on each characteristic vector to obtain clustering result data;

[0048] Screening the reaction features according to the clustering result data to obtain target features, and determining feature weight values ​​of each target feature;

[0049] A fault detection result of the battery cell is determined according to the feature weight value of each target feature and the feature vector of each target feature.

[0050] The above-mentioned battery fault detection method, device, computer equipment, storage medium and computer program product obtain the electrochemical impedance spectrum data of the battery cell and generate an electrochemical reaction spectrum based on multiple frequency points in the electrochemical impedance spectrum data, as well as the real impedance data and imaginary impedance data of each frequency point. The use of real and imaginary data to jointly generate the electrochemical reaction spectrum can ensure information integrity, make full use of the electrochemical impedance response in complex form, and mine the typical behavior patterns at multiple time scales inside the battery, thereby overcoming the problem of missing reaction details caused by relying solely on the imaginary impedance. Next, by calculating the characteristic vectors of multiple reaction features in the electrochemical reaction spectrum and performing cluster analysis on each characteristic vector, clustering result data is obtained. The clustering result data can identify the distribution differences of different reaction features between samples, thereby eliminating the problems of strong subjectivity and weak robustness when manually screening features, and improving the processing capabilities of weak signals or boundary distribution data. Next, the reaction features are screened according to the clustering result data to obtain the target features, and the feature weight value of each target feature is determined; the fault detection result of the battery cell is determined according to the feature weight value of each target feature and the feature vector of each target feature, and the fault detection result of the battery is determined using the target feature vector with weights, realizing a detection process that does not require model priors, feature self-screening, and self-weighting, thereby effectively improving the accuracy of battery fault identification in small samples or boundary scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0052] Figure 1 A diagram illustrating an application environment of a battery failure detection method according to an embodiment;

[0053] Figure 2 1 is a flow chart of a battery fault detection method according to an embodiment;

[0054] Figure 3 1 is a flow chart of step S208 of a battery failure detection method according to an embodiment;

[0055] Figure 4 is a structural block diagram of a battery fault detection device in one embodiment;

[0056] Figure 5 is a diagram of the internal structure of a computer device in one embodiment;

[0057] Figure 6 FIG. 4 is a diagram showing the internal structure of a computer device in another embodiment. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0059] The battery fault detection method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network, and the terminal 102 can be used to obtain the electrochemical impedance spectroscopy data of the battery cell and send the data to the server 104. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The data storage system can be used to store the electrochemical impedance spectroscopy data of the battery cell, as well as historical battery data, and can also be used to store relevant data of the kernel density estimation model constructed based on historical battery data. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablets, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 104 can be implemented as an independent server or a server cluster consisting of multiple servers.

[0060] In an exemplary embodiment, Figure 2 As shown, a battery fault detection method is provided, which is applied to Figure 1 The server 104 in the example is used as an example to illustrate the method, which includes the following steps S202 to S208.

[0061] Step S202 : acquiring electrochemical impedance spectroscopy data of the battery cell, and generating an electrochemical reaction spectrum according to multiple frequency points in the electrochemical impedance spectroscopy data, and real impedance data and imaginary impedance data of each frequency point.

[0062] Among them, electrochemical impedance spectroscopy (EIS) data refers to the response of a battery to a small AC signal applied to the battery at different frequencies. It is a frequency characteristic response data used to reflect the dynamic behavior of the battery, such as charge transfer, electrode process, electrolyte diffusion, etc. Electrochemical impedance spectroscopy data can be composed of multiple frequency points, each frequency point corresponds to a complex impedance value, and the complex impedance Z is composed of the real impedance data Re(Z) and the imaginary impedance data Im(Z). The real part is used to represent the dissipation behavior of energy, such as resistive loss, and the imaginary part is used to represent the phenomenon of temporary storage or delayed release of energy in the interface process or double layer capacitance. When collecting EIS data, the server 104 can preset a frequency range, such as from 10kHz to 10mHz, and collect dozens to hundreds of frequency points within this range. For example, at a frequency of 1000Hz, the battery may measure Re(Z) of 0.3Ω and Im(Z) of -0.1Ω, while at 0.1Hz, Re(Z) may reach 5Ω and Im(Z) of -1.2Ω.

[0063] The electrochemical impedance spectroscopy (EIS) data may be collected by the terminal 102 and the server 104 in a coordinated manner.

[0064] For example, in some embodiments, the terminal 102 can independently complete the impedance measurement, and the terminal 102 can be deployed in a position close to the battery, such as embedding a measurement module in the battery management system, or being equipped with a portable EIS acquisition device. The terminal 102 actively collects the impedance response data of the battery at a series of frequency points locally on a regular basis or on demand. After the collection is completed, the real and imaginary parts of the impedance corresponding to each frequency point are uploaded to the server 104 through the communication interface. It is suitable for scenarios where the data reporting frequency is not high, the network communication is stable, or the terminal has a certain edge processing capability. For example, in a certain power battery monitoring scenario, the terminal 102 can scan the current battery status every 4 hours, collect the impedance data of 100 frequency points, and then package the data packet and send it to the server 104, and can attach metadata such as timestamp and battery number.

[0065] For example, in other embodiments, the terminal 102 can also collect EIS data according to the collection instruction sent by the server 104 and send it to the server 104. The server 104 can send the collection instruction through control signaling based on factors such as the system's detection cycle, task scheduling, and fault prediction requests. The instruction may include the frequency range, number of sampling points, and measurement time window to be collected, and may also include the polling order of multiple battery channels. After receiving the instruction, the terminal 102 can start the corresponding EIS measurement process and send the collected data back to the server in a timely manner. For example, when the server 104 finds that a battery pack has a trend drift after analyzing a set of historical data, it can immediately send a high-frequency sampling instruction to the terminal 102, requesting a more refined EIS scan to be performed on the battery and quickly update its reaction spectrum.

[0066] Next, the server 104 can use these frequency points as the horizontal axis coordinates and the real and imaginary impedances as input information to jointly derive an electrochemical reaction spectrum reflecting the multi-time constant reaction mechanism of the battery, wherein the electrochemical reaction spectrum refers to the distribution of relaxation times (DRT), which can be expressed as a functional relationship between the relaxation time constant τ and the corresponding intensity γ, reflecting the distribution of various internal processes of the battery at different time scales. In this embodiment, the spectrum constructed by the server 104 can not adopt the traditional simplified fitting method based only on imaginary part information, but instead simultaneously introduces real and imaginary part data for joint fitting to improve the stability of the solution and the physical interpretability of the spectrum.

[0067] For example, the server 104 may first map the real and imaginary data into two structure matrices, which respectively simulate the theoretical contributions of different time constants to the real and imaginary impedances; then, the server 104 may transform the problem into an optimization problem with regularization constraints, ensuring that the spectrum solution has good fitting performance while retaining the sparsity and physical interpretability of the spectrum. The formula is as follows:

[0068]

[0069] Where A1 and A2 are the real and imaginary kernel matrices, respectively, generated by discretizing the DRT integral equation, and α is the joint regularization parameter. For example, if a battery has typical charge transfer and diffusion processes, the spectrum may have distinct peaks near τ = 1ms and τ = 10s. Server 104 can automatically locate these peaks using an optimization algorithm and output the spectrum data in the form of a one-dimensional sparse vector.

[0070] For example, the server 104 may adopt an augmented Lagrangian solution and introduce an alternating direction method of multipliers (ADMM) algorithm to iteratively solve the convex optimization problem with constraints, so as to efficiently solve the objective function with sparse constraints. After receiving the impedance spectrum data transmitted by the terminal 102, the server 104 may organize the real part data and the imaginary part data into two vectors respectively, and calculate the relaxation time constant based on a preset range (e.g., from 10-3 to 10 2seconds), construct the corresponding kernel matrix. These two kernel matrices describe the projection contribution of the reaction processes with different time constants in the real and imaginary parts of the impedance, respectively, and are the mathematical bridge for mapping the DRT distribution to the observable impedance value. On this basis, the server 104 defines the solution goal as: to find a non-negative DRT vector γ′ so that after passing through the kernel matrix, it can restore the measured Re(Z) and Im(Z) to the greatest extent, while maintaining sparseness in the L1 norm sense, that is, to retain only a few important reaction processes as much as possible. The server 104 can use the ADMM method to update the block variables.

[0071] For example, the server 104 uses the target vector γ′ as the primary variable and initializes its estimated value. In each iteration, the server 104 can update the primary variable γ′ based on the current residual. This step aims to minimize the fitting error and fit the original impedance data as closely as possible. The following formula is used in the variable update step:

[0072]

[0073] Subsequently, the server 104 can map the primary variable to an auxiliary variable space and introduce a soft threshold function in the space to implement sparse constraints. Through this process, the server 104 automatically removes small peaks close to zero and retains only the response process with significant intensity, ensuring that the final map is concise and clear. The auxiliary variable update adopts the following formula:

[0074] z k+1 =soft-threshold α / ρ (γ′ k+1 +u k )

[0075] Among them, the soft threshold function implements L1 sparsification:

[0076] soft-threshold λ (x)=sign(x)max(|x|-λ,0)

[0077] Finally, server 104 updates the dual variables to maintain numerical stability of the solution between the variables, so that the algorithm does not diverge or oscillate. The above three stages are executed cyclically until the difference between the main variable and the auxiliary variable and the residual between the two iterations are both lower than the preset threshold, indicating that the solution process has stabilized and converged. The dual variable update uses the following formula:

[0078] u k+1 =u k +γ′ k+1 -z k+1

[0079] The server 104 can also adopt an automatic optimization mechanism of the regularization parameter α in this process. In order to balance the fitting accuracy and sparsity, the server 104 can divide the sample set into a training set and a validation set, calculate the average error when fitting the real part and the imaginary part respectively, and construct a Pareto front graph based on this. The server 104 selects an error equilibrium point as the optimal α value based on the graph to ensure that the obtained DRT spectrum will neither have overfitting peaks nor lose important reaction information due to excessive regularization constraints. The server 104 can adopt joint optimization of regularization parameters, use dual-objective cross-validation to determine the optimal α, divide the training set / validation set, and calculate the independent fitting errors of the real part and the imaginary part:

[0080]

[0081] Construct the Pareto frontier and select α that minimizes Ere+EIm:

[0082] α opt =argmin α (w1E Re +w2E Im ),w1+w2=1

[0083] Ultimately, the DRT vector γ′ that the server 104 can output will show several clear non-zero peaks, and the position of each peak corresponds to a relaxation time constant with physical meaning (for example, representing a diffusion process or charge transfer process), and its amplitude reflects the weight of the process in the overall impedance response.

[0084] In the above steps, the server 104 does not rely on a preset equivalent circuit model or manual judgment, but instead inverts the internal response structure of the battery based on the data itself to obtain an electrochemical reaction spectrum, so that subsequent feature extraction can be based on a more complete and objective physical mechanism from the source.

[0085] Step S204 , calculating characteristic vectors of multiple reaction features in the electrochemical reaction spectrum, and performing cluster analysis on each characteristic vector to obtain clustering result data.

[0086] Exemplarily, the server 104 can perform a traversal analysis on each of the above-mentioned generated DRT maps, and extract multiple statistical features as the basic dimensions of the feature vector. These statistical features include but are not limited to: entropy (Shannon entropy), mean, standard deviation, skewness and kurtosis, which describe the performance of the DRT map in terms of numerical form, distribution structure and extreme changes from different angles. Shannon entropy is used to measure the degree of confusion of the map information; the mean is used to reflect the overall reaction intensity; the standard deviation is used to indicate the degree of dispersion of each point in the map; the skewness is used to evaluate whether the map distribution is symmetrical, and the kurtosis is used to describe whether there is a sharp concentration or extremely abnormal structure. The server 104 can extract these five indicators one by one for the DRT map of each battery cell to form an original five-dimensional feature vector, and use the feature set formed by multiple samples as input to enter the next stage of cluster analysis.

[0087] During the cluster analysis process, the server 104 can adopt a one-dimensional clustering strategy based on the Gaussian Mixture Model (GMM). Unlike traditional K-means, GMM allows each cluster to present a probability distribution structure, which is suitable for handling complex situations such as overlapping and asymmetric feature distributions. The server 104 does not cluster the entire five-dimensional vector as a whole, but performs one-dimensional GMM modeling analysis on each feature dimension separately. Therefore, the server 104 can independently determine whether each feature has the ability to cluster the samples, and thus determine whether this feature really produces a distinguishable change between different types of batteries.

[0088] Furthermore, in some embodiments, the features extracted by server 104 are as follows:

[0089] (1) Entropy (Shannon entropy). An increase in entropy indicates an increase in the complexity of the relaxation process:

[0090]

[0091] (2) Average value, used to measure the overall intensity level of DRT:

[0092]

[0093] (3) Standard deviation, used to characterize the degree of dispersion of the relaxation process:

[0094]

[0095] (4) Skewness, distribution asymmetry:

[0096]

[0097] (5) Kurtosis, existence of extreme relaxation processes:

[0098]

[0099] Step S206 , screening the reaction features according to the clustering result data to obtain target features, and determining the feature weight value of each target feature.

[0100] Exemplarily, the server 104 may determine the number of clusters for each feature vector based on the clustering result data; when the number of clusters is greater than or equal to the screening threshold, the corresponding reaction feature is used as the target feature; the intra-class dispersion and inter-class dispersion of each target feature are calculated based on the clustering result data; and the feature weight value of each target feature is calculated based on the intra-class dispersion and inter-class dispersion.

[0101] The number of clusters is used to represent the number of distribution structures of the reaction characteristics naturally separated among samples, which can be the number of Gaussian components fitted in the above GMM model. Perform GMM clustering independently:

[0102]

[0103] where K is determined dynamically by the Bayesian Information Criterion:

[0104] BIC=-2lnL+Klnn

[0105] For example, the server 104 may set a screening threshold, such as a cluster number of not less than 2, indicating that the feature can at least divide the samples into two different groups and has preliminary distinguishing ability. For features with a cluster number less than the threshold (i.e., K < 2), for example, if a feature is almost uniformly distributed across all samples, the GMM can only fit a Gaussian component. In this case, the server 104 may identify the feature as an undistinguishable feature and remove it from use as a target feature. However, for features with a cluster number not less than the threshold, the server 104 may include them as target features.

[0106] Taking the aforementioned "kurtosis" feature as an example, server 104 can input the numerical values ​​of all samples along this dimension into the GMM module and fit a Gaussian mixture model along this dimension using maximum likelihood estimation. Simultaneously, server 104 can use the Bayesian Information Criterion to evaluate the model fit under different numbers of mixture components to select the optimal number of clusters, K. If the feature clearly forms two or more Gaussian components across samples (i.e., K ≥ 2), server 104 can deem the feature to have effective clustering capabilities and select the corresponding response feature as the target feature. However, if the fitting results indicate that the feature only forms a single Gaussian distribution (i.e., K = 1), indicating that it cannot effectively distinguish between samples, server 104 can discard the feature to prevent it from interfering with subsequent diagnostic procedures. In this process, server 104 not only retains high-value feature dimensions that can distinguish normal from abnormal batteries but also initially completes a coarse screening mechanism for feature validity. Server 104 avoids the instability and high computational complexity associated with direct clustering in high-dimensional space and also makes the subsequent weighting process more targeted. The server 104 may retain the clustering result of each valid feature as a cluster label and its probability value, and use these label data as intermediate variables to prepare for the next stage of feature screening and weighting.

[0107] Next, the server 104 can further analyze the intra-class dispersion and inter-class dispersion of each feature. Among them, the intra-class dispersion refers to the fluctuation size of the feature within each cluster. If the numerical difference of a feature in the same class of samples is very small, it means that it has good consistency in the class and is a stability indicator. On the contrary, if the intra-class variance is too large, it means that the feature cannot be clearly expressed within the group, and the indicator can be considered unreliable. The inter-class dispersion is used to represent the mean difference between different cluster centers. The greater the difference, the more the feature can distance different classes and has a stronger discriminant power. The server 104 can generate a discriminant score for each target feature by calculating the ratio of the square of the inter-class mean difference (as a discrimination index) to the intra-class variance (as a stability index) of each target feature, that is, the Fisher discriminant criterion.

[0108] For example, the server 104 may calculate the intra-class dispersion of the i-th feature in the k-th class:

[0109]

[0110] Calculate the inter-class dispersion (the difference between the global mean μ and the class mean μk):

[0111]

[0112] The higher the score of the target feature, the more obvious the difference between different sample types, but the more stable and reliable it is within the same type. Server 104 normalizes the Fisher scores of all target features and finally assigns a standardized weight value to each target feature. This weight value will directly determine the influence ratio of the feature in subsequent feature fusion, principal component analysis, trend judgment and other operations. Fisher score and weight distribution:

[0113] (m is the total number of features retained)

[0114] For example, if the three target features are skewness, kurtosis, and entropy, and their Fisher discriminant scores are 8.5, 21.2, and 5.3, respectively, server 104 can normalize these three features to: skewness 0.22, kurtosis 0.55, and entropy 0.23. When subsequently using weighted principal component analysis or constructing a weighted composite index, the diagnostic role of kurtosis will be prominently enhanced.

[0115] Through this step, server 104 no longer relies on human experience or heuristic thresholds to determine feature importance. Instead, it relies entirely on the actual distribution characteristics of the sample data, using clustering structures and automatic quantification of statistics. Especially when the sample size is small or the sample structure is complex, this cluster distribution-driven feature screening mechanism can effectively avoid noise accumulation caused by redundant features and improve the robustness of subsequent judgment models.

[0116] Step S208 : determining a fault detection result of the battery cell according to the feature weight value of each target feature and the feature vector of each target feature.

[0117] For example, server 104 can construct a globally discriminative fused feature representation based on the target feature vector of each battery cell and its corresponding weight. Server 104 can weightedly sum all target feature values ​​or convert them into a set of low-dimensional principal components, such as PC1 and PC2, through weighted principal component analysis, as a compressed representation of the overall state of the battery at the current moment.

[0118] After obtaining the fused state vector, the server 104 can map the battery to a pre-constructed health feature distribution space to assess whether it deviates from the normal range. This space can be constructed by kernel density estimation, which can be established in the historical step using normal battery sample data. The server 104 can calculate the outlier factor of the battery based on its density value in the distribution, and use this as a basis to determine the fault level. For example, if the outlier factor is lower than the threshold θcore, it is judged to be healthy; if the outlier factor is between θcore and 1.5 times θcore, it is a mild abnormality; if the outlier factor exceeds 1.5 times θcore, it is a serious abnormality. The result is the current fault score data of the battery, which will be immediately recorded and output to the upper-level application or operation and maintenance personnel. In addition, the server 104 can also establish a fault feature library (such as the characteristic change pattern corresponding to internal short circuit, aging, and lithium precipitation) through historical data, and use the nearest neighbor classification to match the current outlier.

[0119] For example, after a battery cell completes fault detection, server 104 may also add the fault score data of the battery cell to an update set; if the number of battery cells included in the update set is greater than or equal to an update threshold, the fault threshold is updated based on the fault score data of each battery cell in the update set; and the update set is cleared. This enables the solution to have adaptive update capabilities to cope with distribution drift that occurs as the battery population status evolves over time.

[0120] Furthermore, after a battery completes a fault detection, the server 104 can include its outlier factor and fused feature vector in an update set for temporary data caching. The update set is used to continuously collect newly generated test results in the recent period as a basis for the system to perform self-calibration of the judgment criteria. Once the cumulative number of battery samples in the update set reaches a preset update threshold (for example, every 50 battery cells), the server 104 can trigger a fault judgment threshold update process. In the incremental threshold update process, after each k new samples are added, the threshold is updated based on the latest data distribution:

[0121]

[0122] where a is a smoothing coefficient (default 0.2), and Q0.95 is the 95th percentile. In the updating process, the server 104 can statistically analyze all the Degree of Outlier Factor (DOF) data to be updated, re-estimate the 95th percentile in the normal sample distribution, and calculate the new core and the serious abnormal threshold value accordingly. The server 104 can also introduce an exponential smoothing method so that the new threshold value does not fluctuate sharply. After the update is completed, the original fault determination limit can be automatically replaced by the new threshold value, thereby ensuring that the system still has the ability to adapt continuously even if the battery type is replaced, the environmental temperature changes, or the aging process advances. After the update is completed, the server 104 can immediately empty the update set to prepare for the next round of data accumulation.

[0123] Exemplarily, assuming that the DOF value of a certain battery sample is 18 after the target feature vector fusion, and the current system core threshold value is 12, the server 104 determines that the battery has a slight fault, and adds the DOF to the update set. The current update set already has 49 data, and this one will constitute the 50th, triggering the server 104 to re-calculate the distribution boundary of all DOF values, and find that the new core should be adjusted to 13.4, that is, it is updated and continues the subsequent fault determination task. Through this feedback type closed loop updating mechanism, the server 104 realizes the smooth transition from static rules to self-learning judgment model, effectively solving the problem of rule invalidation and scene drift in long-term operation.

[0124] In the above-mentioned battery fault detection method, the electrochemical impedance spectroscopy data of the battery cell is obtained, and the electrochemical reaction spectrum is generated based on multiple frequency points in the electrochemical impedance spectroscopy data, as well as the real impedance data and imaginary impedance data of each frequency point. The use of real and imaginary data to jointly generate the electrochemical reaction spectrum can ensure information integrity, make full use of the electrochemical impedance response in complex form, and mine the typical behavior patterns at multiple time scales within the battery, thereby overcoming the problem of missing reaction details caused by relying solely on the imaginary impedance. Next, by calculating the characteristic vectors of multiple reaction features in the electrochemical reaction spectrum and performing cluster analysis on each characteristic vector, clustering result data is obtained. The clustering result data can identify the distribution differences of different reaction features between samples, thereby eliminating the problems of strong subjectivity and weak robustness when manually screening features, and improving the processing capabilities of weak signals or boundary distribution data. Next, the reaction features are screened according to the clustering result data to obtain the target features, and the feature weight value of each target feature is determined; the fault detection result of the battery cell is determined according to the feature weight value of each target feature and the feature vector of each target feature, and the fault detection result of the battery is determined using the target feature vector with weights, realizing a detection process that does not require model priors, feature self-screening, and self-weighting, thereby effectively improving the accuracy of battery fault identification in small samples or boundary scenarios.

[0125] In an exemplary embodiment, Figure 3 As shown, step S208 includes steps S302 to S306.

[0126] Step S302: Acquire historical battery data.

[0127] For example, server 104 can further introduce historical battery data as a reference to evaluate the relative position and deviation of the current battery state in the context of a larger time scale and population distribution. By comparing the difference in fused features between the target features of the battery cell to be tested and historical normal samples, it can be quantified whether it is abnormal and the degree of abnormality. Server 104 can load a batch of confirmed historical battery data from local storage or a cloud database, or directly obtain the kernel density estimation model established using normal battery sample data in the historical step.

[0128] Historical battery data can include two parts: the feature vector data of each battery, and the status label of this data at the time of collection, such as "healthy," "early failure," or "severe abnormality." These historical battery samples can cover multiple collection cycles and various operating conditions, representing the true distribution of normal battery performance in the system.

[0129] Step S304 : Calculate the fault score data of the battery cell according to the feature vector, feature weight value and historical battery data of each target feature.

[0130] Subsequently, server 104 can combine the feature vector of the target feature of the current battery with the existing feature weight values ​​to perform a weighted fusion operation to generate the final fused feature representation of the battery. For example, if the three target features of the current sample are skewness = 0.7, kurtosis = 2.5, and entropy = 0.95, and the weights are 0.25, 0.5, and 0.25, respectively, server 104 can construct the principal component by summing these weights to obtain the fused feature vector [PC1 = 1.675, PC2 = 0.26]. This result is essentially a coordinate position of the battery in the multidimensional feature space.

[0131] Server 104 then substitutes this fused feature point into the fault scoring model and calculates its "fault scoring data," which is the degree of outlier (DOF) of the point under the historical normal sample distribution. This can be expressed as the inverse of the probability density of the battery under the historical sample density function. If the sample is in a low-density area in the historical healthy sample distribution, f(x) is small, and DOF = 1 / f(x) will become significantly larger, indicating a significant deviation from the battery's past characteristics.

[0132] Exemplarily, the server 104 may perform dimensionality reduction on the feature vector using a weighted principal component analysis method based on the feature weight values ​​of each target feature to obtain a first principal component vector and a second principal component vector; the feature weight values ​​corresponding to each target feature in the first principal component vector are greater than the feature weight values ​​corresponding to each target feature in the second principal component vector; a time series feature vector is constructed based on the first principal component vector and historical battery data; and the fault score data of the battery cell is calculated based on the time series feature vector, the first principal component vector, and the second principal component vector.

[0133] Furthermore, the server 104 may input the time series feature vector, the first principal component vector, and the second principal component vector into a kernel density estimation model to obtain a probability density value; and calculate the fault score data of the battery cell according to the probability density value.

[0134] The kernel density estimation model is constructed based on the characteristic data of historical non-faulty battery cells. Server 104 can perform weighted principal component analysis (WPCA) based on the weight values ​​of each target feature to reduce the dimensionality of the high-dimensional feature space and construct a comprehensive feature vector that best represents the battery status. In this process, server 104 can simultaneously model the current battery status and evolution trend by retaining the dominant component and appropriately constructing the secondary principal components, thereby enhancing the fault identification ability to perceive potential changes.

[0135] For example, before starting the dimensionality reduction operation, the server 104 may first reweight the target feature vector of each battery according to its corresponding feature weight to form a weighted feature matrix. w =[w1x1,w2x2,...,w m x m ] T For example, if the three target features of a battery are skewness, kurtosis, and entropy, and their corresponding weights are 0.2, 0.6, and 0.2, respectively, then the feature vector of the sample will be adjusted to a weighted matrix consisting of the original value multiplied by the weight.

[0136] Next, the server 104 may perform covariance analysis on the weighted matrix and perform eigenvalue decomposition The two most representative principal components are extracted and named the first principal component vector and the second principal component vector. The first principal component vector aggregates the one or two target features with the largest weights, reflecting the main variation direction of the battery health status. The second principal component captures possible minor abnormal fluctuations while maintaining the data variation structure, supporting fault trend judgment. For example, the first principal component PC1 retains the first two features with the largest Fisher weights:

[0137]

[0138] The second principal component PC2 integrates the remaining features:

[0139]

[0140] For example, if the weights for skewness, kurtosis, and entropy are 0.2, 0.6, and 0.2, respectively, server 104 may find during analysis that the first principal component is primarily driven by kurtosis, with a coefficient of 0.7, while the coefficients for skewness and entropy are 0.2 and 0.1, respectively. This indicates that this principal component best represents health changes driven by sharp electrochemical response characteristics. In contrast, the weights for skewness and entropy in the second principal component may be relatively even, at 0.5 and 0.5, respectively, while the kurtosis is only between 0.0 and 0.1, indicating that this component is more sensitive to changes in information complexity and asymmetry.

[0141] After obtaining the first and second principal components, the server 104 can construct a time series feature vector based on the first principal component vector and the corresponding principal component time series in the historical battery data. Use the historical data to construct global statistics and extract the global mean drift:

[0142]

[0143] Among them, μ historyis the PC1 mean of historical normal data. Then extract the global coefficient of variation ratio:

[0144]

[0145] By comparing the coefficient of variation of the current batch with historical normal data, the overall distribution change is quantified, the feature space is reconstructed, and a four-dimensional fusion feature vector is constructed:

[0146] X fusion =[PC1,PC2,Δμ,CV] T

[0147] This vector is used to measure the degree of change in the current battery status in the long-term trend, that is, whether the current principal component value has shifted compared to the historical mean, and whether its volatility has increased significantly. Server 104 can calculate the difference Δμ between the current first principal component mean and the historical mean as the average offset, and compare the coefficient of variation of the current and historical samples to obtain the CV ratio as the volatility enhancement. These two indicators are combined to form a time series feature, which is used to enhance the fault identification perception of the "evolution" of the battery status, rather than just making a static judgment on the current snapshot.

[0148] Finally, the server 104 combines the first principal component vector, the second principal component vector, and the time series feature vector constructed above to form a complete four-dimensional feature expression: [PC1, PC2, Δμ, CV ratio]. The server 104 can input this four-dimensional vector into the kernel density estimation model previously trained based on historical normal samples, and calculate the probability density f(x) of the battery in the multidimensional feature space using the Gaussian kernel function. For example, a fixed-bandwidth kernel density estimation (KDE) is used to calculate the global density distribution:

[0149]

[0150] The bandwidth matrix h is automatically determined by the Silverman criterion:

[0151]

[0152] in, is the sample standard deviation. If the density value is high, it means that the current battery is still within the healthy distribution range; if the density value is low, it means that its behavior characteristics have deviated significantly from the normal sample, and there is a potential risk.

[0153] In order to achieve specific quantification of the abnormality degree, the server 104 can further calculate the outlier factor DOF=1 / f(x) as the fault score data. The degree to which a battery deviates from its normal state is expressed numerically, with values ​​closer to 1 indicating a higher degree of outlier. For example, if the principal component eigenvector of a battery is [1.4, 0.3, 0.25, 1.5], and server 104 inputs this into the KDE model and calculates a density value f(x) = 0.02, then DOF = 50. Server 104 then compares this DOF ​​with the fault threshold θcore. For example, if θcore = 15, the sample has significantly exceeded the normal range, and server 104 determines it as "severely abnormal" and immediately issues an alert.

[0154] Step S306 : When the fault score data exceeds the fault threshold, it is determined that the battery cell is in a fault state.

[0155] For example, server 104 can determine a core determination threshold θcore based on the DOF distribution statistically obtained from historical healthy battery samples. This threshold can be set to the 95th percentile of the DOF of normal samples, representing the maximum allowable deviation from the healthy samples. For example, if server 104 has historically stored DOF data for 1,000 healthy battery samples and, after sorting, finds that the 950th DOF value is 14.8, θcore is set to 14.8. This value represents the most acceptable abnormality; DOF values ​​below this threshold are considered within the normal fluctuation range.

[0156] Next, server 104 can compare the current battery cell's DOF ​​value with θcore and enter a grading judgment process. If the battery's DOF ​​is lower than θcore, server 104 can directly determine it to be healthy. If its DOF ​​is between θcore and 1.5 times θcore, server 104 can determine it to be "incipient fault," indicating that the battery's characteristics have statistically shifted slightly but persistently, such as a slightly larger Δμ or a slightly increased CV ratio, but no structural changes have occurred. Once the battery's DOF ​​value exceeds 1.5θcore, server 104 can mark it as "critical fault," deeming its characteristics to have significantly deviated and classified as a high-risk device. For example, if θcore = 14.8, a battery with a DOF of 12 is considered normal, a battery with a DOF of 20 is considered an incipient fault, and a battery with a DOF of 35 is considered a critical fault.

[0157] In addition, the feature vectors of different types of failed batteries and their corresponding DOF distributions can also be recorded in the historical failure samples maintained by the server 104. The server 104 can calculate a lower threshold θ outlier from these samples, which can be the 5th percentile of the historical failure DOF. This value represents the mildest known failure behavior. In the case where the DOF of the current sample exceeds θ outlier, it enters the failure mode mapping process to further confirm whether the anomaly is consistent with the performance characteristics of a certain known failure type.

[0158] In terms of failure mode mapping, the server 104 can construct a feature library containing typical failure modes in advance. This library can be formed by modeling the feature changes (especially Δμ and CV ratio) of different failure samples such as internal short circuit, aging, lithium dendrite precipitation, etc. determined in the past. Each type of failure can exhibit a specific distribution pattern in this feature space, for example, the Δμ of aging type failure usually rises slowly but the CV ratio is stable, while the CV ratio of internal short circuit type failure abnormally rises and the Δμ fluctuates dramatically.

[0159] The server 104 can compare the [Δμ, CV ratio] vector of the currently failed battery sample with all templates in the feature library one by one, using K-neighbor matching method to find the closest failure template to the current feature. If the time series features of this battery are similar to multiple samples in the internal short circuit template, the server 104 can determine that it may have an internal short circuit risk, and output this determination result as a specific failure label. For example, if the current battery Δμ = 0.42, CV ratio = 2.1, and the lithium dendrite precipitation type sample is concentrated in the range of Δμ = 0.40.5, CV ratio = 2.02.3, then the server 104 can classify this sample as a lithium dendrite precipitation type failure through K-neighbor classification.

[0160] The dual threshold judgment mechanism of θ core and θ outlier of the server 104 not only realizes the accurate judgment of whether a failure occurs, but also effectively distinguishes the early stage and severe stage of the failure, and completes the intelligent mapping recognition of the failure type based on the pre-trained feature template library. This mechanism can not only guarantee the sensitivity and accuracy of failure judgment, but also avoid false positives and misjudgments, and is particularly suitable for deployment in large-scale battery group scenarios as an important support module for health management and intelligent scheduling.

[0161] In the above embodiment, server 104 utilizes a process involving weighted principal component analysis, time series feature construction, kernel density estimation, and DOF calculation. This not only achieves intelligent dimensionality reduction from high-dimensional features to a compressed structure, but also establishes a fusion evaluation mechanism that integrates static states and dynamic trends, resulting in fault scoring results with enhanced sensitivity, interpretability, and temporal consistency. This mechanism is particularly suitable for complex battery behavior, where early anomalies are difficult to identify with a single metric, effectively improving the ability to identify edge cases and hidden faults.

[0162] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0163] Based on the same inventive concept, embodiments of the present application also provide a battery fault detection device for implementing the aforementioned battery fault detection method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more battery fault detection device embodiments provided below can be found in the above-described limitations of the battery fault detection method and will not be further elaborated here.

[0164] In an exemplary embodiment, Figure 4 As shown, a battery fault detection device is provided, including: a data acquisition module 402, a feature clustering module 404, a feature screening module 406 and a fault detection module 408, wherein:

[0165] The data acquisition module 402 is used to acquire electrochemical impedance spectroscopy data of the battery cell and generate an electrochemical reaction spectrum based on multiple frequency points in the electrochemical impedance spectroscopy data and the real impedance data and imaginary impedance data of each frequency point;

[0166] The feature clustering module 404 is used to calculate the feature vectors of multiple reaction features in the electrochemical reaction spectrum and perform cluster analysis on each feature vector to obtain clustering result data;

[0167] The feature screening module 406 is used to screen the reaction features according to the clustering result data, obtain target features, and determine the feature weight value of each target feature;

[0168] The fault detection module 408 is configured to determine a fault detection result of the battery cell according to a feature weight value of each target feature and a feature vector of each target feature.

[0169] In one embodiment, the feature screening module 406 is specifically used to: determine the number of clusters for each feature vector based on the clustering result data; when the number of clusters is greater than or equal to the screening threshold, use the corresponding reaction feature as the target feature; calculate the intra-class dispersion and inter-class dispersion of each target feature based on the clustering result data; and calculate the feature weight value of each target feature based on the intra-class dispersion and inter-class dispersion.

[0170] In one embodiment, the fault detection module 408 includes:

[0171] A data acquisition unit, used to acquire historical battery data;

[0172] A data analysis unit, configured to calculate the fault score data of the battery cell based on the feature vector, feature weight value and historical battery data of each target feature;

[0173] The fault detection unit is used to determine that the battery cell is in a fault state when the fault score data exceeds the fault threshold.

[0174] In one embodiment, the data analysis unit is specifically used to: reduce the dimension of the feature vector using the weighted principal component analysis method according to the feature weight value of each target feature to obtain a first principal component vector and a second principal component vector; the feature weight value corresponding to each target feature in the first principal component vector is greater than the feature weight value corresponding to each target feature in the second principal component vector; construct a time series feature vector based on the first principal component vector and historical battery data; and calculate the fault score data of the battery cell based on the time series feature vector, the first principal component vector and the second principal component vector.

[0175] In one embodiment, the data analysis unit is specifically used to: input the time series feature vector, the first principal component vector and the second principal component vector into a kernel density estimation model to obtain a probability density value; wherein the kernel density estimation model is constructed based on the characteristic data of historical non-faulty battery cells; and calculate the fault score data of the battery cell based on the probability density value.

[0176] In one embodiment, the device further includes: a threshold update module, configured to add the fault score data of the battery cell to an update set after the battery cell completes fault detection; when the number of battery cells included in the update set is greater than or equal to the update threshold, updating the fault threshold based on the fault score data of each battery cell in the update set; and clearing the update set.

[0177] The modules in the battery fault detection apparatus can be implemented by software, hardware, or a combination thereof. The modules can be embedded in or independent of a processor in a computer device in hardware form, or stored in a memory in the computer device in software form, so as to be invoked and executed by the processor to perform the operations corresponding to the modules.

[0178] In an exemplary embodiment, a computer device, which can be a server, is provided, and an internal structure diagram of the computer device can be as shown in Figure 5 The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store electrochemical impedance spectrum data of battery cells, historical battery data, and related data of a kernel density estimation model constructed according to the historical battery data. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with terminals outside through a network connection. The computer program is executed by the processor to implement a battery fault detection method.

[0179] In an exemplary embodiment, a computer device, which can be a terminal, is provided, and an internal structure diagram of the computer device can be as shown in Figure 6As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface, the display unit and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a battery fault detection method is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse.

[0180] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0181] In an exemplary embodiment, a computer device is provided, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented: obtaining electrochemical impedance spectrum data of a battery cell, and generating an electrochemical reaction spectrum based on multiple frequency points in the electrochemical impedance spectrum data, as well as real impedance data and imaginary impedance data of each frequency point; calculating characteristic vectors of multiple reaction features in the electrochemical reaction spectrum, and performing cluster analysis on each characteristic vector to obtain clustering result data; screening the reaction features based on the clustering result data to obtain target features, and determining a characteristic weight value for each target feature; and determining a fault detection result for the battery cell based on the characteristic weight value and the characteristic vector of each target feature.

[0182] In one embodiment, when the processor executes the computer program, the following steps are further implemented: determining the number of clusters for each feature vector based on the clustering result data; when the number of clusters is greater than or equal to the screening threshold, using the corresponding reaction feature as the target feature; calculating the intra-class dispersion and inter-class dispersion of each target feature based on the clustering result data; and calculating the feature weight value of each target feature based on the intra-class dispersion and inter-class dispersion.

[0183] In one embodiment, when the processor executes the computer program, the processor further implements the following steps: obtaining historical battery data; calculating fault score data of the battery cell based on the feature vector, feature weight value and historical battery data of each target feature; and determining that the battery cell is in a fault state when the fault score data exceeds a fault threshold.

[0184] In one embodiment, when the processor executes the computer program, the following steps are further implemented: based on the feature weight value of each target feature, the feature vector is reduced in dimension using a weighted principal component analysis method to obtain a first principal component vector and a second principal component vector; the feature weight value corresponding to each target feature in the first principal component vector is greater than the feature weight value corresponding to each target feature in the second principal component vector; based on the first principal component vector and historical battery data, a time series feature vector is constructed; and based on the time series feature vector, the first principal component vector and the second principal component vector, fault score data of the battery cell is calculated.

[0185] In one embodiment, when the processor executes the computer program, the following steps are further implemented: inputting the time series feature vector, the first principal component vector, and the second principal component vector into a kernel density estimation model to obtain a probability density value; wherein the kernel density estimation model is constructed based on the characteristic data of historical non-faulty battery cells; and calculating the fault score data of the battery cell based on the probability density value.

[0186] In one embodiment, when the processor executes the computer program, the following steps are further implemented: after the battery cell completes fault detection, the fault score data of the battery cell is added to the update set; when the number of battery cells included in the update set is greater than or equal to the update threshold, the fault threshold is updated according to the fault score data of each battery cell in the update set; and the update set is cleared.

[0187] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: obtaining electrochemical impedance spectroscopy data of a battery cell, and generating an electrochemical reaction spectrum based on multiple frequency points in the electrochemical impedance spectroscopy data, as well as real impedance data and imaginary impedance data at each frequency point; calculating characteristic vectors of multiple reaction features in the electrochemical reaction spectrum, and performing cluster analysis on each characteristic vector to obtain clustering result data; screening the reaction features based on the clustering result data to obtain target features, and determining a characteristic weight value for each target feature; and determining a fault detection result for the battery cell based on the characteristic weight value and the characteristic vector of each target feature.

[0188] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: determining the number of clusters for each feature vector based on the clustering result data; when the number of clusters is greater than or equal to a screening threshold, using the corresponding reaction feature as a target feature; calculating the intra-class dispersion and inter-class dispersion of each target feature based on the clustering result data; and calculating the feature weight value of each target feature based on the intra-class dispersion and inter-class dispersion.

[0189] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: obtaining historical battery data; calculating fault score data of the battery cell based on the feature vector, feature weight value and historical battery data of each target feature; and determining that the battery cell is in a fault state when the fault score data exceeds a fault threshold.

[0190] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: based on the feature weight value of each target feature, the feature vector is reduced in dimension using the weighted principal component analysis method to obtain a first principal component vector and a second principal component vector; the feature weight value corresponding to each target feature in the first principal component vector is greater than the feature weight value corresponding to each target feature in the second principal component vector; based on the first principal component vector and historical battery data, a time series feature vector is constructed; and based on the time series feature vector, the first principal component vector and the second principal component vector, the fault score data of the battery cell is calculated.

[0191] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: inputting the time series feature vector, the first principal component vector, and the second principal component vector into a kernel density estimation model to obtain a probability density value; wherein the kernel density estimation model is constructed based on the characteristic data of historical non-faulty battery cells; and calculating the fault score data of the battery cell based on the probability density value.

[0192] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: after the battery cell completes fault detection, the fault score data of the battery cell is added to the update set; when the number of battery cells included in the update set is greater than or equal to the update threshold, the fault threshold is updated according to the fault score data of each battery cell in the update set; and the update set is cleared.

[0193] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0194] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided in this application may be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, etc., but are not limited to these.

[0195] Any technical features in the above embodiments can be combined, and for the sake of brevity, not all possible combinations are described above, however, as long as the combinations of technical features do not have contradictions, they shall be considered within the scope of the present disclosure.

[0196] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it shall not be understood as a limitation on the patent scope of the present application. It shall be pointed out that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and these shall be within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A battery fault detection method, characterized in that: The method comprises: Acquire electrochemical impedance spectroscopy data of the battery cell, and generate an electrochemical reaction spectrum based on multiple frequency points in the electrochemical impedance spectroscopy data, as well as real impedance data and imaginary impedance data at each frequency point; Calculating characteristic vectors of a plurality of reaction features in the electrochemical reaction spectrum, and performing cluster analysis on each of the characteristic vectors to obtain clustering result data; Screening the reaction features according to the clustering result data to obtain target features, and determining feature weight values ​​of each target feature; A fault detection result of the battery cell is determined according to the feature weight value of each target feature and the feature vector of each target feature.

2. The method according to claim 1, characterized in that The step of screening the reaction features according to the clustering result data to obtain target features and determining feature weight values ​​of the target features includes: Determining the number of clusters for each of the feature vectors according to the clustering result data; When the number of clusters is greater than or equal to a screening threshold, the corresponding reaction feature is used as a target feature; Calculating the intra-class dispersion and inter-class dispersion of each target feature according to the clustering result data; The feature weight value of each target feature is calculated according to the intra-class dispersion and the inter-class dispersion.

3. The method according to claim 1, characterized in that The determining the fault detection result of the battery cell according to the feature weight value of each target feature and the feature vector of each target feature includes: Get historical battery data; Calculating fault score data of the battery cell according to the feature vector of each target feature, the feature weight value and the historical battery data; When the fault score data exceeds a fault threshold, it is determined that the battery cell is in a fault state.

4. The method according to claim 3, characterized in that The calculating the fault score data of the battery cell according to the feature vector of each target feature, the feature weight value and the historical battery data includes: According to the feature weight values ​​of each of the target features, the feature vector is subjected to dimensionality reduction using a weighted principal component analysis method to obtain a first principal component vector and a second principal component vector; the feature weight values ​​corresponding to each of the target features in the first principal component vector are greater than the feature weight values ​​corresponding to each of the target features in the second principal component vector; constructing a time series feature vector based on the first principal component vector and the historical battery data; Fault score data of the battery cell is calculated according to the time series feature vector, the first principal component vector, and the second principal component vector.

5. The method according to claim 4, characterized in that The calculating the fault score data of the battery cell according to the time series feature vector, the first principal component vector, and the second principal component vector includes: Inputting the time series feature vector, the first principal component vector, and the second principal component vector into a kernel density estimation model to obtain a probability density value; wherein the kernel density estimation model is constructed based on characteristic data of historical non-faulty battery cells; Fault score data of the battery cell is calculated according to the probability density value.

6. The method according to any one of claims 3 to 5, characterized in that The method further comprises: After the battery cell completes fault detection, adding the fault score data of the battery cell to an update set; When the number of the battery cells included in the update set is greater than or equal to an update threshold, updating the fault threshold according to the fault score data of each battery cell in the update set; Clear the update set.

7. A battery fault detection device, characterized in that: The device comprises: A data acquisition module, configured to acquire electrochemical impedance spectroscopy data of a battery cell and generate an electrochemical reaction spectrum based on multiple frequency points in the electrochemical impedance spectroscopy data and the real impedance data and imaginary impedance data of each frequency point; A feature clustering module, configured to calculate feature vectors of a plurality of reaction features in the electrochemical reaction spectrum, and perform cluster analysis on each feature vector to obtain clustering result data; A feature screening module is used to screen the reaction features according to the clustering result data to obtain target features and determine the feature weight value of each target feature; A fault detection module is configured to determine a fault detection result of the battery cell according to the feature weight value of each target feature and the feature vector of each target feature.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images