A battery anomaly detection method based on chebyshev polynomials

Abstract

The application relates to a battery abnormality detection method based on Chebyshev polynomials and belongs to the technical field of battery management. The method comprises the following steps: S1, collecting electric vehicle operation data, including battery pack voltage data and other parameters of each battery monomer, and performing data preprocessing; S2, performing Kalman filtering processing on the preprocessed battery monomer voltage data; and dividing the filtered data into a training set and a verification set; S3, based on the filtered battery monomer voltage data, performing Chebyshev polynomial feature extraction, and extracting polynomial coefficients as features; S4, using the Chebyshev polynomial coefficients to reconstruct the approximate value of the original data, and calculating the reconstruction error; S5, setting an abnormality detection threshold, and judging whether each monomer battery is faulty according to whether the reconstruction error exceeds the set threshold. The method can effectively realize electric vehicle power battery abnormality detection and accurately locate the abnormal monomer.

Description

Technical Field

[0001] This invention belongs to the field of battery management technology and relates to a battery anomaly detection method based on Chebyshev polynomials. Background Technology

[0002] With the increasing popularity of electric vehicles (EVs), the performance, safety, and lifespan of the power battery pack, as the core energy storage unit of EVs, directly determine the vehicle's range, reliability, and operating costs. Therefore, establishing a precise, rapid, and reliable battery fault diagnosis system is of paramount importance for ensuring the safe operation of EVs, preventing catastrophic accidents such as thermal runaway, and extending the lifespan of the power battery.

[0003] Currently, the mainstream battery fault detection technologies in the industry can be mainly divided into two categories: threshold judgment methods based on direct measurement parameters and analysis methods based on battery models.

[0004] The first type of method directly monitors key physical quantities of battery cells, such as voltage, current, and temperature, and compares them with preset fixed thresholds. When the measured values ​​exceed safe ranges, the system identifies a fault. While this method is simple, intuitive, and easy to implement, its limitations are significant. It can only identify faults that have progressed to a relatively severe stage, and it struggles to provide effective early warnings for early-stage, subtle potential faults. Furthermore, the fixed thresholds cannot adapt to the dynamically changing normal operating windows of batteries at different life stages, operating temperatures, and charge / discharge rates, easily leading to false alarms or missed alarms.

[0005] The second type of approach involves establishing a mathematical model of the battery and diagnosing faults by comparing the residuals between the model's predicted values ​​and the actual sensor measurements. Among these, analytical methods based on equivalent circuit models (ECMs) and electrochemical models are a hot research topic. However, these traditional model-based methods face numerous challenges in practical applications. For example, using numerical methods such as the finite difference method (FDM) to solve complex nonlinear partial differential equations to construct a high-precision electrochemical model is extremely cumbersome. Especially when dealing with complex boundary conditions and geometric configurations, the computational load is enormous, making it difficult to meet the stringent real-time requirements of vehicle management systems (VMS). More importantly, numerical methods such as the finite difference method are sensitive to data noise. In the complex electromagnetic and vibration environments of actual electric vehicle operation, noise inevitably gets mixed into the signals collected by sensors, which severely interferes with the model's calculation process and significantly reduces the accuracy and reliability of fault detection.

[0006] Furthermore, most existing Battery Management Systems (BMS) often use a uniform, idealized battery model to represent all cells in the entire battery pack when performing fault diagnosis. This "one-size-fits-all" approach severely ignores the inherent inconsistencies between individual cells within the battery pack. Due to minute differences in manufacturing processes, uneven temperature distribution during long-term use, and different aging rates, the electrochemical characteristics (such as internal resistance, capacity, and open-circuit voltage) of each battery cell exhibit unique, dynamically changing differences over time. Applying a standard model to all cells inevitably masks subtle anomalies caused by early faults that resemble normal individual differences. This makes it difficult to adapt to the diverse diagnostic needs of batteries of different types and aging stages, ultimately leading to insufficient accuracy and sensitivity in diagnosis.

[0007] Finally, traditional fault detection methods generally lack robustness when faced with the complex and ever-changing real-world operating conditions of electric vehicles (such as rapid acceleration, rapid deceleration, and bumpy road conditions) and strong noise interference. These external disturbances can cause drastic fluctuations in battery parameters, making it difficult for traditional methods to effectively distinguish them from actual internal battery faults (such as internal short circuits, loose connections, and electrolyte leakage). This frequently leads to false positives and false negatives, reducing system reliability and user trust.

[0008] In summary, existing battery fault detection methods suffer from significant bottlenecks in terms of diagnostic accuracy, real-time performance, adaptability to individual differences, and robustness under complex operating conditions. To overcome these shortcomings, there is an urgent need to develop a novel battery fault detection method that can effectively handle the highly nonlinear characteristics of battery systems, intelligently adapt to individual differences between battery cells, and maintain high stability and reliability even in complex noise environments, in order to meet the higher safety and reliability requirements of next-generation electric vehicles. Summary of the Invention

[0009] In view of this, the purpose of this invention is to provide a battery anomaly detection method based on Chebyshev polynomials. By fitting battery voltage data and extracting Chebyshev polynomial coefficients as features, the data is reconstructed using these features and the reconstruction error is calculated, thereby achieving anomaly detection. This method not only reduces the impact of noise and improves detection sensitivity, but also balances detection accuracy and computational complexity by adjusting the polynomial order, adapting to different battery characteristics and operating conditions, and significantly improving the accuracy and robustness of battery fault detection.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] A battery anomaly detection method based on Chebyshev polynomials specifically includes the following steps:

[0012] S1: Collect electric vehicle operation data, including voltage data of each battery cell in the battery pack and other parameters, and perform data preprocessing;

[0013] S2: Perform Kalman filtering on the preprocessed battery cell voltage data to reduce noise interference; and divide the filtered data into training set and validation set to provide a data foundation for subsequent feature extraction and anomaly detection.

[0014] S3: Based on the filtered battery cell voltage data, Chebyshev polynomial feature extraction is performed, and the polynomial coefficients are extracted as features.

[0015] S4: Reconstruct approximate values ​​of the original data using Chebyshev polynomial coefficients and calculate the reconstruction error;

[0016] S5: Set an anomaly detection threshold. Based on whether the reconstruction error exceeds the set threshold, determine whether each individual cell has malfunctioned.

[0017] Furthermore, step S1 specifically includes the following steps:

[0018] S11: Collect normal operation data of a certain electric vehicle, including parameters such as total voltage, voltage of each individual battery cell, and time;

[0019] S12: Perform data filtering and cleaning on the extracted battery cell voltages, and fill in missing values ​​in the data;

[0020] S13: Establish a power battery operation database based on the collected operating data of the power battery pack and individual battery cells.

[0021] Furthermore, in step S2, the preprocessed battery cell voltage data is subjected to Kalman filtering, specifically including the following steps:

[0022] S21: Based on the preprocessed individual unit voltage data, determine the initial state estimate and the estimation error covariance;

[0023] S22: Use the mean and variance of the first few data points as the initial state estimate, use the state estimate of the previous time step to predict the state at the current time step, and update the estimation error covariance;

[0024] S23: Calculate the Kalman gain based on the current measurement, correct the prediction state to obtain a more accurate estimate, and update the estimation error covariance;

[0025] S24: Filter the voltage time series of each battery cell, and output a smooth and stable voltage data series at each time point.

[0026] Furthermore, in step S2, the filtered data is divided into a training set and a validation set. Specifically, this includes: splitting the filtered data sorted by time into the first 70% of normal data and the last 30% of abnormal data; based on the split data, the first 70% of normal data is defined as the training set, and the last 30% of abnormal data is defined as the validation set, providing a data foundation for subsequent feature extraction and anomaly detection.

[0027] In the data partitioning method, the validation set can be based on normal data or abnormal data, and the partitioning rules are optional.

[0028] Furthermore, step S3 specifically includes the following steps:

[0029] S31: Based on experience or through cross-validation, select an appropriate Chebyshev polynomial order m to balance the accuracy and computational complexity of feature extraction; choosing a higher Chebyshev polynomial order can capture more detailed features, but may also lead to overfitting.

[0030] To reduce the impact of abnormal individual cells on the average voltage, the normalized median voltage value is compared with the normalized individual cell voltage values ​​to achieve battery fault detection.

[0031] S32: Construct Chebyshev polynomial basis functions;

[0032] Define m+1 Chebyshev polynomial basis functions T j (x), where j = 0, 1, ..., m;

[0033] Satisfies the recurrence relation:

[0034] T0(x)=1

[0035] T1(x)=x

[0036] T j+1 (x)=2xT j (x)-T j-1 (x), j≥1

[0037] Where x represents the independent variable, which is the number corresponding to the zeros of the Chebyshev polynomial mapped from the battery cell voltage data, and is an equally spaced point on the interval [-1, 1]; m represents the order of the Chebyshev polynomial; j represents the order index of the Chebyshev polynomial, which is a non-negative integer (j≥0) used to identify different Chebyshev polynomials; T jT(x) represents the j-th Chebyshev polynomial, a function of x used to represent Chebyshev polynomials of different orders. T0(x) is the 0-th order Chebyshev polynomial, T1(x) is the 1-th order Chebyshev polynomial, and T... j+1 (x) is a Chebyshev polynomial of order j+1, T j-1 (x) is a Chebyshev polynomial of order j-1;

[0038] S33: Based on Chebyshev polynomial basis functions T j For each individual cell in the training and validation sets, the voltage data is fitted using the least squares method to obtain the polynomial coefficients c. j This minimizes the sum of squared errors between the fitted curve and the original data; the extracted coefficients c j As a characteristic of individual battery cell voltage data, it reflects the main trends and patterns of change in the data.

[0039] Furthermore, step S4 specifically includes the following steps:

[0040] S41: Using the extracted Chebyshev polynomial coefficients c j and basis functions T j (x), reconstructing an approximation of the original data.

[0041]

[0042] in, x represents the approximate value of the reconstructed original data for the i-th data point; i T represents the i-th independent variable; j (x i ) represents the j-th order Chebyshev polynomial in x. i The value at;

[0043] S42: Compare with the original data y i and reconstructing data Calculate reconstruction error

[0044] Furthermore, step S5 specifically includes the following steps:

[0045] S51: Analyze the reconstruction error distribution of the validation set, select an appropriate threshold setting method, such as fixed threshold method, dynamic threshold method or threshold method based on statistical distribution, and determine the anomaly detection threshold.

[0046] S52: The reconstruction error e of the validation set i Compare with the threshold; if e i If the threshold is greater than 1, then the data of a single battery cell at that moment is considered an outlier.

[0047] S53: If a battery malfunctions, the faulty cell will be located based on its voltage.

[0048] The beneficial effects of this invention are as follows:

[0049] 1) This invention utilizes Chebyshev polynomial feature extraction to effectively separate different frequency components of data by leveraging its orthogonality, thereby extracting more representative and independent features. This not only enhances the distinguishing ability of the features but also significantly improves the accuracy of anomaly detection, enabling the method to more sensitively capture subtle changes in battery voltage data.

[0050] 2) The excellent convergence speed of the method of the present invention enables it to achieve high-precision data approximation at a lower order, thereby reducing the number of required features and reducing the consumption of computing resources. This is especially important in large-scale battery monitoring scenarios, as it can shorten the detection time and improve the system response speed.

[0051] 3) This invention has good robustness to noise and can stably extract key features of battery voltage in noisy data environments. This ensures that the method can still provide reliable results when facing complex working conditions and noise interference in practical applications, reducing false alarms and missed alarms and enhancing the overall stability of the system.

[0052] 4) The battery anomaly detection method proposed in this invention has high flexibility and wide applicability, adapting to various battery types and complex operating conditions. This makes the method suitable for many fields such as electric vehicles and energy storage systems. At the same time, its high sensitivity and robustness enable it to detect battery anomalies in a timely and accurate manner, providing strong support for battery safety management and maintenance.

[0053] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0054] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0055] Figure 1 The overall flowchart of the battery anomaly detection method based on Chebyshev polynomials provided in Example 1;

[0056] Figure 2 This is a diagram of the algorithm framework for Example 1;

[0057] Figure 3 The individual cell voltage curves for normal and faulty cells in Example 2;

[0058] Figure 4 The Kalman filter curves of the voltage of the normal and faulty cells in Example 2 are shown.

[0059] Figure 5 The feature extraction results based on Chebyshev polynomials in Example 2;

[0060] Figure 6 This is a comparison chart of the original training data and the reconstructed data in Example 2;

[0061] Figure 7 This is a comparison chart of the original data and reconstructed data of the validation set in Example 2;

[0062] Figure 8 This is the training set reconstruction error map in Example 2;

[0063] Figure 9 This is the reconstruction error map of the validation set in Example 2;

[0064] Figure 10 This is the result of the validation set anomaly detection in Example 2. Detailed Implementation

[0065] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0066] Example 1:

[0067] This embodiment provides a battery anomaly detection method based on Chebyshev polynomials, which can determine the health status of each cell in the power battery pack based on on-site data of electric vehicles and perform real-time fault detection on each battery cell.

[0068] Please see Figures 1-2 The method specifically consists of the following steps:

[0069] Step S1: Collect electric vehicle operating data, including the voltage data of each cell in the battery pack and other characteristics, and perform data preprocessing, specifically including the following steps:

[0070] Step S11: Collect normal operation data of a certain electric vehicle, including parameters such as total voltage, voltage of each individual cell, and time;

[0071] Step S12: Filter and clean the extracted unit voltage data, and fill in the missing values ​​in the data;

[0072] Data cleaning involves removing rows with a total voltage of 0, and missing data values ​​can be filled using the average value interpolation method between adjacent cells.

[0073] Step S13: Establish a power battery operation database based on the collected operating data of the power battery pack and individual battery cells.

[0074] Step S2: Perform Kalman filtering on the battery cell voltage data to reduce noise interference. This includes the following steps:

[0075] Step S21: Based on the processed unit voltage data, determine the initial state estimate and the estimation error covariance, and the initial state estimate value. The calculation of the error covariance P0 usually takes the mean of the first few data points, and the calculation of the error covariance P0 usually takes the variance of the first few data points.

[0076] Step S22: Use the mean and variance of the first few data points as the initial state estimate, and use the state estimate value of the previous time step k-1. Predict the prior state estimate at time k. And update the prior estimate error covariance P k|k-1 ;

[0077] The prior state estimate at time k The calculation formula is as follows:

[0078]

[0079] Updated prior estimate error covariance P k|k-1 The calculation formula is as follows:

[0080] P k|k-1 =P k-1|k-1 +Q

[0081] Among them, P k-1|k-1 Let Q represent the estimation error covariance at the previous time k-1; Q is the process noise covariance.

[0082] Step S23: Based on the current measured value z k Calculate the Kalman gain K k Correct the predicted state to obtain the posterior state estimate. And update the posterior estimation error covariance P. k|k ;

[0083] Kalman gain K k The calculation formula is as follows:

[0084]

[0085] Where R is the measurement noise covariance.

[0086] Posterior state estimate The calculation formula is as follows:

[0087]

[0088] Posterior estimation error covariance P k|k The calculation formula is as follows:

[0089] P k|k = (1-K) k )P k|k-1

[0090] Step S24: Perform steps S21 to S23 sequentially for the voltage time series of each battery cell, and output a smooth and stable voltage data sequence at each time point.

[0091] Step S3: Based on the filtered unit voltage data, perform Chebyshev polynomial feature extraction, extracting the polynomial coefficients as features. This specifically includes the following steps:

[0092] Step S31: Divide the filtered data sorted by time into the first 70% of normal data and the last 30% of abnormal data;

[0093] Step S32: Based on the segmented data, define the first 70% of normal data as the training set and the last 30% of abnormal data as the validation set, providing a data foundation for subsequent feature extraction and anomaly detection.

[0094] In the data partitioning method, the validation set can be based on normal data or abnormal data, and the partitioning rules are optional.

[0095] Step S4: Based on the filtered unit voltage data, Chebyshev polynomial feature extraction is performed, and the polynomial coefficients are extracted as features. This specifically includes the following steps:

[0096] Step S41: Based on experience or through cross-validation, select an appropriate Chebyshev polynomial order m to balance the accuracy and computational complexity of feature extraction.

[0097] To reduce the impact of abnormal individual cells on the average voltage, the normalized median voltage value is compared with the normalized individual cell voltage values ​​to achieve battery fault detection.

[0098] Step S42: Construct Chebyshev orthogonal polynomial basis functions.

[0099] Define m+1 Chebyshev polynomial basis functions T j (x), where j = 0, 1, ..., m;

[0100] Satisfies the recurrence relation:

[0101] T0(x)=1

[0102] T1(x)=x

[0103] T j+1 (x)=2xT j (x)-T j-1 (x) (j≥1)

[0104] Where x represents the independent variable, that is, the number corresponding to the zeros of the Chebyshev polynomial mapped to the battery voltage data, which are equally spaced points on the interval [-1, 1]; m represents the order of the Chebyshev polynomial; j represents the order index of the Chebyshev polynomial, which is a non-negative integer (j≥0) used to identify different Chebyshev polynomials; T j T(x) represents the j-th Chebyshev polynomial, a function of x used to represent Chebyshev polynomials of different orders. T0(x) is the 0-th order Chebyshev polynomial, T1(x) is the 1-th order Chebyshev polynomial, and T... j+1 (x) is a Chebyshev polynomial of order j+1, T j-1 (x) is a Chebyshev polynomial of order j-1.

[0105] Step S43: Based on the Chebyshev orthogonal polynomial basis function T j For each individual cell in the training and validation sets, the voltage data is fitted using the least squares method to obtain the polynomial coefficients c. j This minimizes the sum of squared errors between the fitted curve and the original data. The extracted coefficients c j As a characteristic of battery voltage data, it reflects the main trends and patterns of change in the data.

[0106] Step S5: Reconstruct approximate values ​​of the original data using Chebyshev polynomial coefficients and calculate the reconstruction error. This includes the following steps:

[0107] Step S51: Utilize the extracted Chebyshev polynomial coefficients c j and basis functions T j (x), reconstructing an approximation of the original data.

[0108]

[0109] in, x represents the approximate value of the reconstructed original data for the i-th data point; i T represents the i-th independent variable x; j (x i ) represents the j-th order Chebyshev polynomial in x. i The value at that location.

[0110] Step S52: Compare with the original data y i and reconstructing data Calculate reconstruction error

[0111] Step S6: Set an anomaly detection threshold. Based on whether the reconstruction error exceeds the set threshold, determine whether each individual cell has failed. This specifically includes the following steps:

[0112] Step S61: Analyze the reconstruction error distribution of the validation set, select an appropriate threshold setting method, such as fixed threshold method, dynamic threshold method or threshold method based on statistical distribution, and determine the anomaly detection threshold.

[0113] Step S62: Calculate the reconstruction error e of the validation set. i Compare with the threshold; if e i If the threshold is greater than 1, then the data for that individual battery cell at that moment is considered an outlier.

[0114] Step S63: If the battery malfunctions, locate the faulty cell based on the voltage of the faulty cell.

[0115] The location of the faulty unit includes the faulty unit number and the location of the abnormal point.

[0116] Example 2:

[0117] To illustrate the effectiveness of the present invention, this embodiment provides the operating data of a faulty vehicle, in which a battery cell malfunctions and triggers an alarm from the BMS. Figure 3 This section presents the individual cell voltage curves for normal and faulty cells in the battery of the faulty vehicle. Kalman filtering and normalization were then applied to the data. Figure 4 Kalman filter curves of the voltage of normal and faulty cells in the battery of the faulty vehicle are used. The data is divided into a training set (first 70% of normal data) and a validation set (last 30% of abnormal data). A Chebyshev polynomial of order m = 10 is selected. Figure 5 For feature extraction results, low-order features reflect the main trends in the data, while high-order features may capture subtle variations. Chebyshev polynomials and preserved coefficients are used to reconstruct approximations of the original data. Figure 6To compare the original training data with the reconstructed data, Figure 7 To compare the original data with the reconstructed data in the validation set, anomalies are detected using the reconstruction error of the validation set, and the reconstruction error of the training set is observed, with a fixed threshold of 0.335 set. Figure 8 For the reconstruction error of the training set, Figure 9 The value represents the reconstruction error on the validation set. The mean squared error (MSE) of both the training and validation sets is less than 0.2, and the reconstruction error of the training set is less than 0.335. Figure 10 To verify the anomaly detection results, no false alarms were detected for normal cells, and faulty cells and anomalies were effectively identified. Therefore, this invention can effectively extract the main features of the data, reduce noise interference, and capture subtle changes in the data, enabling better real-time fault detection of electric vehicle power batteries and accurate location of faulty cells.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A battery anomaly detection method based on Chebyshev polynomials, characterized in that, The method specifically includes the following steps: S1: Collect electric vehicle operation data, including voltage data of each battery cell in the battery pack and other parameters, and perform data preprocessing; S2: Perform Kalman filtering on the preprocessed battery cell voltage data; and divide the filtered data into training and validation sets. S3: Based on the filtered battery cell voltage data, Chebyshev polynomial feature extraction is performed, and the polynomial coefficients are extracted as features; specifically, the following steps are included: S31: Choose the Chebyshev polynomial order based on experience or through cross-validation. To balance the accuracy and computational complexity of feature extraction; S32: Construct Chebyshev polynomial basis functions; definition Chebyshev polynomial basis functions ,in, ; Satisfies the recurrence relation: in, The independent variable is the number that maps the battery cell voltage data to the zeros of the Chebyshev polynomial, which is within the interval. Equally spaced points on; This represents the order of the Chebyshev polynomial; The order index of a Chebyshev polynomial is a non-negative integer used to identify different Chebyshev polynomials; Indicates the first Chebyshev polynomials are a set of polynomials about The function is used to represent Chebyshev polynomials of different orders. It is a Chebyshev polynomial of order 0. It is a first-order Chebyshev polynomial. yes Chebyshev polynomials yes Chebyshev polynomial; S33: Based on Chebyshev polynomial basis functions For the voltage data of each individual cell in the training and validation sets, a Chebyshev polynomial is fitted using the least squares method to obtain the polynomial coefficients. This minimizes the sum of squared errors between the fitted curve and the original data; the extracted coefficients As a characteristic of individual battery cell voltage data, it reflects the trend and change pattern of the data; S4: Reconstruct approximate values ​​of the original data using Chebyshev polynomial coefficients and calculate the reconstruction error; specifically including the following steps: S41: Utilizing the extracted Chebyshev polynomial coefficients and basis functions Reconstructing approximate values ​​of the original data : in, Indicates the first The approximate value of reconstructing the original data for each data point; Indicates the first One independent variable; Indicates the first Chebyshev polynomials in The value at; S42: Compare the original data and reconstructing data Calculate reconstruction error ; S5: Set an anomaly detection threshold. Based on whether the reconstruction error exceeds the set threshold, determine whether each individual cell has malfunctioned.

2. The battery anomaly detection method based on Chebyshev polynomials according to claim 1, characterized in that, Step S1 specifically includes the following steps: S11: Collect normal operating data for a certain electric vehicle, including total voltage, individual battery cell voltage, and time; S12: Perform data filtering and cleaning on the extracted battery cell voltages, and fill in missing values ​​in the data; S13: Establish a power battery operation database based on the collected operating data of the power battery pack and individual battery cells.

3. The battery anomaly detection method based on Chebyshev polynomials according to claim 1, characterized in that, In step S2, the preprocessed battery cell voltage data is subjected to Kalman filtering, which specifically includes the following steps: S21: Based on the preprocessed individual unit voltage data, determine the initial state estimate and the estimation error covariance; S22: Use the mean and variance of the first few data points as the initial state estimate, use the state estimate of the previous time step to predict the state at the current time step, and update the estimation error covariance; S23: Calculate the Kalman gain based on the current measurement, correct the prediction state to obtain a more accurate estimate, and update the estimation error covariance; S24: Filter the voltage time series of each battery cell, and output a smooth and stable voltage data series at each time point.

4. The battery anomaly detection method based on Chebyshev polynomials according to claim 1, characterized in that, In step S2, the filtered data is divided into a training set and a validation set. Specifically, the filtered data sorted by time is split into the first 70% of normal data and the last 30% of abnormal data. Based on the split data, the first 70% of normal data is defined as the training set, and the last 30% of abnormal data is defined as the validation set.

5. The battery anomaly detection method based on Chebyshev polynomials according to claim 1, characterized in that, Step S5 specifically includes the following steps: S51: Analyze the reconstruction error distribution of the validation set, select a threshold setting method, and determine the anomaly detection threshold. ; S52: Reconstruction error of the validation set With threshold If a comparison is made, Then determine the first i The individual battery cell data points are outliers; S53: If a battery malfunctions, the faulty cell will be located based on its voltage.

6. The battery anomaly detection method based on Chebyshev polynomials according to claim 5, characterized in that, In step S51, the selected threshold setting method includes fixed threshold method, dynamic threshold method or threshold method based on statistical distribution.

Images