A Model-Driven Data Fusion Approach for Micro-Short Circuit and Low-Capacity Fault Diagnosis and Classification in Energy Storage Lithium-ion Batteries

Abstract

This invention discloses a model- and data fusion-driven method for diagnosing and classifying micro-short circuit and low-capacity faults in energy storage lithium batteries, belonging to the field of battery management systems. The method includes: estimating the battery's ohmic internal resistance by combining a first-order integer equivalent circuit model and a Rint model; and classifying faults by combining the battery's relative state of charge (SOC) and relative capacity. By applying the Rint model, effective fault classification can be achieved even when the battery's state of mind is not fully known. The importance of this technique lies in its ability to maintain high fault diagnosis accuracy even with low sampling density, which is significant for battery management and fault early warning in practical applications.

Description

Technical Field

[0001] This invention belongs to the field of battery management systems, and particularly relates to a model and data fusion-driven method for diagnosing and classifying micro-short circuit and low-capacity faults in energy storage lithium batteries for sparse data. Background Technology

[0002] Significant progress has been made in the field of fault diagnosis for photovoltaic energy storage systems and power energy storage systems in recent years. Existing technical frameworks mainly focus on data analysis methods, utilizing large amounts of historical operating data and real-time monitoring data for fault detection and diagnosis. Based on data forming a sample set of equipment operating states, an anomaly judgment mechanism is established by calculating the distance between samples, effectively improving diagnostic efficiency and accuracy.

[0003] However, current methods generally rely on large amounts of labeled samples and high-quality historical data, which not only increases labor costs but also makes it impossible to obtain reliable diagnostic results with small samples or sparse data. Secondly, these methods are insufficiently adaptable to the dynamic changes in energy storage systems. When the system's structure or operating mode changes, models built on historical data often cannot adjust quickly, thus affecting the accuracy of fault detection. Furthermore, existing technologies are weak in detecting latent faults, such as micro-short circuit faults, which can lead to serious consequences, but traditional methods have failed to effectively address their complexity. Finally, many existing technologies also suffer from computational efficiency issues when processing large-scale, complex data. This performance bottleneck often reduces the overall efficiency and real-time performance of fault detection when facing real-time monitoring and diagnosis. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a model- and data fusion-driven method for diagnosing and classifying micro-short circuit and low-capacity faults in energy storage lithium batteries, comprising:

[0005] Obtain the operating data uploaded by the battery management system of industrial and commercial energy storage equipment, perform data cleaning on the operating data, and obtain cleaned data;

[0006] The cleaned data is subjected to feature analysis and selection to obtain a feature dataset;

[0007] The feature dataset is calculated based on the DBSCAN algorithm to obtain the faulty battery. The cluster current and cluster SOC data of the faulty battery are then subjected to data interpolation and data alignment to obtain the interpolated cluster current and cluster SOC information of the BCU.

[0008] The average ohmic internal resistance of the faulty battery was calculated based on a first-order integer-order battery equivalent circuit model.

[0009] Based on the cluster current and cluster SOC information of the interpolated BCU, the voltage information of individual cells collected in the BMU is interpolated, and the average OCV-SOC curve of the normal cell is fitted according to the Rint model and the average ohmic internal resistance.

[0010] The relative SOC change value is calculated based on the average OCV-SOC curve, and the relative capacity is calculated based on the relative SOC change value. At the same time, fault diagnosis and classification thresholds are designed.

[0011] The relative capacity and the fault diagnosis and classification threshold are compared to obtain the diagnosis and classification results of battery MSC and low capacity fault.

[0012] Preferably, the process of calculating the faulty battery based on the DBSCAN algorithm on the feature dataset includes:

[0013] Calculate the number of points in the neighborhood of each data point;

[0014] Randomly select a core point and add it to the first cluster. Continue to add all core points in the neighborhood of the selected core point to the cluster. Repeat this process until all core points are included.

[0015] Select the core points that have not yet been added to a cluster, add them to the next cluster, and repeat the above steps until all core points are included;

[0016] Non-core points that are not added to any cluster are marked as noise points to obtain faulty battery data.

[0017] Preferably, the expression for the first-order integer-order battery equivalent circuit model is:

[0018]

[0019] Among them, U k Battery terminal voltage U t Discrete representation, I k Battery terminal current I t The discrete representation is given by: OCV is the open-circuit voltage of the battery, R0 is the DC internal resistance of the battery, R1 and C1 are the polarization resistor and capacitor, and U1 is the polarization voltage on the RC pair.

[0020] Preferably, the expression for fitting the average OCV-SOC curve of a normal battery based on the Rint model and the average ohmic internal resistance is as follows:

[0021] OCV=U t +I t R0;

[0022] Among them, U t I is the open-circuit voltage of the battery. tThis represents the battery terminal current.

[0023] Preferably, the process of fitting the average OCV-SOC curve of a normal battery based on the Rint model and the average ohmic internal resistance further includes: fitting the calculation results using a multi-order polynomial function, judging the convergence of the fitting by the root mean square error, and screening usable battery data.

[0024] Preferably, the process of calculating the relative SOC change value based on the average OCV-SOC curve, and calculating the relative capacity based on the relative SOC change value, includes:

[0025] Calculate the OCV of the faulty battery, and obtain the SOC change trend of the faulty battery in the SOC range of the normal battery by querying the OCV-SOC data table of the normal battery.

[0026] By calculating the range of SOC changes in the faulty battery, its SOC change amount is obtained. By comparing the SOC change amount of the faulty battery with that of the normal battery, the relative capacity value of the faulty battery is calculated.

[0027] Preferably, the process of comparing the relative capacity and the fault diagnosis with the classification threshold includes:

[0028] Let the charging capacity deviation be q1, the discharging capacity deviation be q2, and the charging / discharging capacity deviation be |q1-q2|=q3;

[0029] Using q1 as the primary criterion for determining low capacity, q3 as the primary criterion for determining MSC, and q2 as the auxiliary criterion for determining MSC, we can distinguish between batteries with only low capacity faults and batteries with both MSC and low capacity faults. The expression is as follows:

[0030]

[0031] Where P1 and P2 are the thresholds for fault diagnosis and classification.

[0032] Preferably, the process of comparing the relative capacity and the fault diagnosis and classification threshold further includes: defining the charge / discharge capacity deviation caused by low capacity as q4 = (q1 + q2) / 2, and the capacity deviation caused by MSC as q5 = (q2 - q1) / 2, and determining the degree of low capacity fault in a battery with both faults simultaneously, expressed as:

[0033]

[0034] Compared with the prior art, the present invention has the following advantages and technical effects:

[0035] 1. Differentiation and diagnosis between micro-short-circuit faults and low-capacity faults: This innovation diagnoses micro-short-circuit faults, which are difficult to diagnose in energy storage systems and can lead to serious accidents, and effectively distinguishes them from easily confused low-capacity faults. This innovation not only improves the accuracy of fault diagnosis but also significantly enhances the safety of energy storage devices, avoiding potential risks caused by misdiagnosis of fault types.

[0036] 2. Simplified and Efficient Unsupervised Learning Framework: A simplified and efficient unsupervised learning framework is adopted. This framework only requires unified data cleaning and processing, eliminating the need for complex annotation and feature extraction operations, greatly reducing the technical complexity and cost of implementation.

[0037] 3. Effective Model and Data Fusion Method for Handling Sparse and Low-Frequency Data: The proposed model and data fusion-driven method can efficiently process sparse and low-frequency sampled data, and is suitable for common data types and characteristics in practical application scenarios. This method takes into account battery mechanisms, effectively classifying battery faults, improving data processing efficiency, and ensuring the accuracy of fault identification. Attached Figure Description

[0038] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0039] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram of the first-order integer-order equivalent circuit model of a battery according to an embodiment of the present invention;

[0041] Figure 3 This is a schematic diagram of the battery equivalent internal resistance model according to an embodiment of the present invention. Detailed Implementation

[0042] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0043] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0044] Example 1

[0045] like Figure 1As shown, this embodiment provides a model and data fusion-driven method for diagnosing and classifying micro-short circuit and low-capacity faults in energy storage lithium batteries, including:

[0046] Step 1: Obtain relevant data of the energy storage battery system based on the cloud platform or other hardware and software platforms, including sampling information of each level of architecture, sampling time, etc.

[0047] Step 2: Perform unified data processing on the obtained information and filter out unusable data.

[0048] Step 3: Merge the battery cell voltage data collected by all BMUs, extract relevant features, and perform feature analysis;

[0049] Step 4: Using the DBSCAN algorithm, based on the selected features, perform data clustering analysis to identify batteries with low-capacity faults and MSC faults;

[0050] Step 5: For BMUs with faulty batteries, perform data interpolation and data alignment based on the battery cluster current and cluster SOC data collected by their respective BCUs to ensure that the voltage collected by the BMUs always has corresponding current information.

[0051] Step 6: Based on the first-order integer-order battery equivalent circuit model, and according to the average voltage and current of the normal batteries in the BMU, estimate the average ohmic internal resistance of the batteries in each BMU.

[0052] Step 7: Based on the cluster current and cluster SOC information of the interpolated BCU, interpolate the individual cell voltage information collected in the BMU, and fit the average OCV-SOC curve of the normal cell according to the Rint model.

[0053] Step 8: Based on the polynomial fitting of the normal battery OCV-SOC curve, establish a normal battery OCV-SOC data table;

[0054] Step 9: Based on the Rint model, the faulty battery voltage, the cluster current of the BCU where the faulty battery is located, and the normal battery OCV-SOC data table, calculate the relative SOC change value of the faulty battery.

[0055] Step 10: Calculate the relative capacity and design fault diagnosis and classification thresholds based on the relative SOC change value;

[0056] Step 11: Output the fault diagnosis results.

[0057] Specifically:

[0058] 1. Data cleaning and extraction

[0059] Because different devices may use different data protocols, valid data (such as battery voltage and current) typically constitutes only a small proportion of the transmitted content. Furthermore, various sensors may be subject to noise interference, leading to outliers or abnormal ranges in the data. Therefore, before officially running the algorithm framework, the uploaded battery data needs to be cleaned and extracted. This step aims to remove invalid or abnormal data, extract key parameters, and ensure the reliability of the algorithm while optimizing the efficiency of computing and storage resource utilization.

[0060] 2. Data alignment and merging

[0061] Commercial and industrial energy storage equipment battery management systems typically employ a three-tier architecture to achieve hierarchical management and control of battery packs, clusters, and stacks. The lowest layer is the Battery Management Unit (BMU), whose main function is to collect data such as voltage and temperature of individual batteries and execute control strategies for those individual batteries. The middle layer is the Battery Cluster Management Unit (BCU), which is primarily responsible for collecting data such as voltage, current, and SOC of the battery cluster, while also aggregating information transmitted from the BMU, communicating with the top layer, and executing cluster-level control commands. The top layer is the Battery Array Management Unit (BAU), responsible for storing and displaying data uploaded by the BMU and communicating and transmitting information with external devices.

[0062] Due to limitations in information transmission and storage costs, the sampling density of individual battery information by the BMU is typically low. Furthermore, the BMU usually employs a polling mechanism when reporting information to the BCU, meaning each BCU queries its subordinate BMUs sequentially at fixed time intervals. This can lead to timestamp misalignment issues in the uploaded battery data. In addition, to meet the high voltage requirements of large-scale commercial and industrial energy storage equipment, the individual battery cells within the BMU and the BMUs under the BCU are usually connected in series. The voltage and temperature data of individual batteries are reported by the BMU, while the battery current and cluster-level average SOC data are stored in the BCU, further exacerbating the timestamp misalignment problem. Therefore, to meet the needs of fault diagnosis, these data require reasonable alignment and integration. Specific alignment and integration principles will be explained in detail in subsequent steps.

[0063] 3. Feature Analysis and Selection

[0064] During discharge, micro-short-circuit (MSC) batteries exhibit similar characteristics to low-capacity batteries, namely, their voltage deviation from that of normal batteries gradually increases. However, during charging, due to the presence of the equipment equalization system, the voltage difference between the faulty battery and the normal battery gradually decreases, regardless of the faulty battery type. Furthermore, MSC faulty batteries are typically accompanied by low-capacity fault characteristics. Based on these characteristics, and considering limitations in data transmission strategies and data quality, this invention utilizes the raw individual battery voltage data uploaded by the BMU for fault identification, selecting the average battery voltage and the difference between the individual battery voltage and the average voltage as the main features.

[0065] Because lithium iron phosphate batteries inevitably exhibit some inconsistency during use, feature analysis based solely on data from a single battery unit (BMU) may lead to misdiagnosis of faults due to minor inconsistencies. Therefore, this approach integrates battery data from all dates for comprehensive analysis. By studying battery voltage data curves, the average voltage and the residual between the individual cell voltage and the average voltage are identified as key features. Considering that the distinction between faulty and normal batteries primarily relies on the residual features, the residual features are multiplied by a coefficient of 2 to enhance their significance, thereby strengthening the clustering effect while reducing the sensitivity and complexity of algorithm parameter adjustments.

[0066] 4. Fault identification based on clustering algorithm

[0067] Commonly used supervised learning methods (such as neural networks or support vector machines) rely on precisely labeled training data, which often requires significant investment of manpower and resources in the early stages of data processing. The data labeling process includes not only data collection and cleaning but also ensuring the accuracy and consistency of labels; otherwise, it may negatively impact model performance. Furthermore, supervised learning methods are highly dependent on the quality and diversity of data labels. If the labels contain errors or the data diversity is insufficient, the model's performance may be unsatisfactory when processing unlabeled or unknown data. In contrast, unsupervised learning methods do not require pre-labeled data but learn by automatically discovering patterns and structures in the data, making them more flexible and adaptable to large-scale and diverse data processing. Although the accuracy of unsupervised learning may be inferior to supervised learning in certain specific tasks, its advantages in exploratory data analysis, data clustering, and dimensionality reduction make it more suitable for the battery fault identification work presented in this paper.

[0068] Data clustering is one of the core methods of unsupervised learning, achieving classification by grouping unordered data into meaningful subclasses. Among numerous clustering algorithms, the DBSCAN algorithm offers greater flexibility when handling unknown datasets because it does not require pre-specifying the number of clusters. Furthermore, the DBSCAN algorithm effectively identifies and handles noise and outliers in the data, marking these points as outliers, thus performing well even with irregular data distributions. However, the selection of parameters for this algorithm is complex and requires precise setting to ensure optimal clustering results.

[0069] The main input parameters of the DBSCAN algorithm include the cluster radius (Eps) and the minimum number of data points (MinPts). Based on these two parameters, the data points in the dataset are divided into three categories: core points, boundary points, and noise points. Core points are data points whose Eps neighborhood contains at least MinPts of points; boundary points are data points whose Eps neighborhood contains fewer than MinPts of points but are located in the neighborhood of core points; noise points are neither core points nor boundary points. The main algorithm flow is as follows:

[0070] (1) Counting the number of points in the neighborhood: Calculate the number of points in the Eps neighborhood for each data point, and mark those with a number greater than MinPts as core points. Neighborhood calculation is based on a specified distance metric; this invention uses Euclidean distance, the formula of which is:

[0071]

[0072] Where x and y represent the feature vectors of the data points; n represents the dimension of the feature.

[0073] (2) Create a cluster: Randomly select a core point, add it to the first cluster, and continue to add all core points in the neighborhood of the core point to the cluster. Repeat this process until all core points are included.

[0074] (3) Add boundary points: Add all non-core points (i.e. boundary points) located in the neighborhood of the core point to the cluster.

[0075] (4) Process the remaining core points: Select the core points that have not yet been added to the cluster, add them to the next cluster, and repeat the above steps until all core points are included.

[0076] (5) Identify noise points: Mark non-core points that have not been added to any cluster as noise points.

[0077] In this invention, the Euclidean distance used by the DBSCAN algorithm can effectively measure the similarity between battery features. By reasonably setting the Eps and MinPts parameters, normal batteries will be classified into the same cluster, while faulty batteries, due to their significant deviation in features, will be identified as noise points, thereby achieving accurate fault identification.

[0078] 5. Battery internal resistance identification

[0079] Because the BMU typically uses a polling mechanism to report information to the BCU, the sampling density of the individual cell voltage reported by the BMU is relatively low, while the sampling density of the cluster current and cluster SOC data reported by the BCU is relatively high. To achieve matching of data with different sampling frequencies, the BCU data is first interpolated. Then, the corresponding cluster current data is matched based on the time point of the BMU data, and a first-order integer model (such as...) is used. Figure 2 As shown, the ohmic internal resistance R0 of the battery is estimated.

[0080] Based on Kirchhoff's laws, the mathematical description of the first-order model is as follows:

[0081]

[0082] Among them U k Battery terminal voltage U t Discrete representation, I k Battery terminal current I t The discrete representation is given by: OCV is the open-circuit voltage of the battery, R0 is the DC internal resistance of the battery, R1 and C1 are the polarization resistor and capacitor, and U1 is the polarization voltage on the RC pair.

[0083] Denote the polarization coefficient Then there is U 1,k =α1U 1,k-1 +R1(1-α1)I k-1 Substitute U 1,k =OCV k -U k -R0I k have:

[0084] α1U 1,k-1 =OCV k -U k -R0I k -R1(1-α1)I k-1 ;

[0085] Formula U 1,k =α1U 1,k-1 +R1(1-α1)I k-1 U in 1,k Change to U 1,k-1 Then we have:

[0086] U 1,k-1 =α1U 1,k-2 +R1(1-α1)I k-2 ;

[0087] Further, there are:

[0088]

[0089] Let equation α1U 1,k-1 =OCV k -U k -R0I k -R1(1-α1)I k-1 U in 1,k-1 Change to U 1,k-2 Then we have:

[0090] α1U 1,k-2 =OCV k-1 -U k-1 -R0I k-1 -R1(1-α1)I k-2 ;

[0091] Further, there are:

[0092] OCV k -U k -R0I k -R1(1-α1)I k-1 =α1[OCV k-1 -U k-1 -R0I k-1 -R1(1-α1)I k-2 ]+α1R1(1-α1)I k-2 ;

[0093] After simplification, we have:

[0094] U k =α1U k-1 +OCV k -α1OCV k-1 -R0I k +α1R0I k-1 -R1(1-α1)I k-1

[0095] In the above formula, U k Change to U k-1 Utilizing the slow time-varying characteristics of OCV, i.e., OCV k -OCV k-1 ≈0, after subtraction we have:

[0096] U k -U k-1 =α1(U k-1 -U k-2 )-R0(I k -I k-1 )-[R1(1-α1)-α1R0](I k-1 -I k-2 );

[0097] Note y k =U k -U k-1 x k =I k -I k-1 , θ1=α1, θ2=-R0, θ3=-[R1(1-α1)-α1R0], then there is:

[0098]

[0099] Where the parameter vector θ = [θ1θ2θ3] T υ k This represents bounded noise and model error.

[0100] The parameter identification of the battery's ohmic internal resistance was completed using the recursive least squares algorithm with a forgetting factor (FFRLS).

[0101]

[0102] in

[0103] 6. Construction of the battery OCV-SOC relationship table

[0104] After estimating the ohmic internal resistance, in order to obtain a smoother and more accurate OCV-SOC curve, the individual cell voltage data of the BMU is further interpolated and aligned using the interpolated BCU data to ensure that the voltage data is synchronized with the cluster SOC and cluster current after the first interpolation.

[0105] Then, using the equivalent internal resistance (Rint) model, based on the battery voltage data after secondary interpolation and the BCU cluster SOC and cluster current data after primary interpolation, the OCV-SOC function of a normal battery is fitted. For example... Figure 3 As shown, the OCV of the Rint model is calculated using the following formula:

[0106] OCV=U t +I t R0;

[0107] Based on the above formula, the OCV is estimated for the longest charging process in each time period and aligned with the SOC. To obtain a smooth and accurate OCV-SOC curve, a multi-order polynomial function is used to fit the calculation results, and the convergence of the fit is judged by the root mean square error (RMSE) to further screen usable battery data.

[0108] Since the fitting function uses a multi-order polynomial, directly calculating the SOC from the OCV of the faulty battery is highly complex. To simplify the SOC estimation process for faulty batteries, this study establishes an OCV-SOC data table based on the fitting function. By looking up the table, the SOC is retrieved from the OCV of the faulty battery, thereby improving the efficiency of SOC estimation.

[0109] 7. Calculation of relative SOC and relative capacity of faulty batteries

[0110] Fault identification has been completed using the DBSCAN algorithm. However, since MSC and low-capacity batteries exhibit similar characteristics during discharge, the key to classification lies in analyzing the SOC changes and capacity calculations during charging. Therefore, it is necessary to further focus on SOC changes during charging and combine this with capacity calculations to distinguish the different characteristics of the two.

[0111] Meanwhile, considering that MSC batteries in real equipment are usually accompanied by low capacity faults and are affected by short-circuit internal resistance, the discharge process of faulty batteries will be greatly affected. Therefore, in the fault identification process, it is necessary to combine the SOC change during the discharge process and the capacity calculation results for auxiliary judgment.

[0112] Therefore, according to OCV=U t +I t R0 calculates the OCV of the faulty battery and obtains the SOC change trend of the faulty battery within the normal battery SOC range by querying the OCV-SOC data table of the normal battery. Furthermore, by calculating the SOC change range of the faulty battery, its SOC change amount is obtained. By comparing the SOC change amounts of the faulty battery and the normal battery, the relative capacity value of the faulty battery is calculated, thereby achieving more accurate fault diagnosis.

[0113] 8. Fault Threshold Design and Classification

[0114] A fault classification threshold was designed by calculating the average charge / discharge capacity and capacity deviation for each faulty battery. Let the charge capacity deviation be q1, the discharge capacity deviation be q2, and the charge / discharge capacity deviation be |q1-q2|=q3. Considering the impact of full-process interpolation and data accuracy, q1 is used as the primary criterion for judging low capacity, q3 as the primary criterion for judging MSC (Mean Cell Missile) fault, and q2 as the auxiliary criterion for judging MSC. According to the following formula, batteries with only low capacity faults and batteries with both MSC and low capacity faults can be distinguished.

[0115]

[0116] P1 and P2 are the thresholds for fault diagnosis and classification, which are set according to the specific conditions of the battery. Typically, P1 = 3% and P2 = 5%.

[0117] Furthermore, for a battery exhibiting both types of faults, the charge / discharge capacity deviation caused by low capacity is defined as q4 = (q1 + q2) / 2, and the capacity deviation caused by MSC is defined as q5 = (q2 - q1) / 2. Based on the following formulas, the degree of the low-capacity fault in a battery exhibiting both faults can be clearly determined.

[0118]

[0119] This completes the diagnosis and classification of battery MSC and low-capacity faults in the energy storage battery system.

[0120] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A model- and data fusion-driven method for diagnosing and classifying micro-short circuit and low-capacity faults in energy storage lithium batteries, characterized in that, include: Obtain the operating data uploaded by the battery management system of industrial and commercial energy storage equipment, perform data cleaning on the operating data, and obtain cleaned data; The cleaned data is subjected to feature analysis and selection to obtain a feature dataset; The feature dataset is calculated based on the DBSCAN algorithm to obtain the faulty battery. The cluster current and cluster SOC data of the faulty battery are then subjected to data interpolation and data alignment to obtain the interpolated cluster current and cluster SOC information of the BCU. The average ohmic internal resistance of the faulty battery was calculated based on a first-order integer-order battery equivalent circuit model. Based on the cluster current and cluster SOC information of the interpolated BCU, the voltage information of individual cells collected in the BMU is interpolated, and the average OCV-SOC curve of the normal cell is fitted according to the Rint model and the average ohmic internal resistance. The relative SOC change value is calculated based on the average OCV-SOC curve, and the relative capacity is calculated based on the relative SOC change value. At the same time, fault diagnosis and classification thresholds are designed. By comparing the relative capacity with the fault diagnosis and classification threshold, the diagnosis and classification results of battery micro-short circuit and low capacity faults are obtained. The process of calculating the relative SOC change based on the average OCV-SOC curve, and calculating the relative capacity based on the relative SOC change, includes: Calculate the OCV of the faulty battery, and obtain the SOC change trend of the faulty battery in the SOC range of the normal battery by querying the OCV-SOC data table of the normal battery. By calculating the range of SOC changes in the faulty battery, its SOC change amount is obtained. By comparing the SOC change amount of the faulty battery with that of the normal battery, the relative capacity value of the faulty battery is calculated. The process of comparing the relative capacity and the fault diagnosis with the classification threshold includes: Let the charging capacity deviation be denoted as The discharge capacity deviation is The charge / discharge capacity deviation is ; by The main criterion for determining low capacity is The main criterion for determining micro short circuits, The auxiliary criterion for determining micro-short circuits is expressed as follows: ； in, , , The threshold is used for fault diagnosis and classification.

2. The method according to claim 1, characterized in that, The expression for fitting the average OCV-SOC curve of a normal battery based on the Rint model and the average ohmic internal resistance is as follows: ； in, This refers to the battery terminal voltage. This is the battery terminal current. is the DC internal resistance of the battery, and OCV is the open-circuit voltage of the battery.

3. The method according to claim 1, characterized in that, The process of fitting the average OCV-SOC curve of a normal battery based on the Rint model and the average ohmic internal resistance further includes: fitting the calculation results with a multi-order polynomial function, judging the convergence of the fitting by the root mean square error, and screening usable battery data.

Images