A data-driven lithium-ion battery fault diagnosis method and system
Through a data-driven method, the multi-scale fuzzy distribution entropy and weighted least squares support vector machine model is used, combined with the improved crown porcupine optimization algorithm, the accurate diagnosis and type identification of lithium-ion battery failures is achieved, and the fault characteristics in the existing technology are solved The problems of insufficient sensitivity and poor real-time performance improve safety.
Patent Information
- Application Number
- CN202411137820.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-19
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-08-19
AI Technical Summary
The existing lithium-ion battery fault diagnosis methods have insufficient sensitivity to fault characteristics, poor real-time performance, and high misjudgment rate, so they cannot accurately and timely determine the fault type, resulting in safety hazards.
Using a data-driven method, by collecting voltage data in electric vehicles, preprocessing and multi-scale fuzzy distribution entropy extraction, a weighted least squares support vector machine model is constructed, combined with the improved crown porcupine optimization algorithm, fault diagnosis and type are realized Identify.
It improves the accuracy and real-time nature of fault diagnosis, reduces the rate of misjudgment, and can accurately identify normal states, sudden failures and gradual failures, adapt to the complex operating conditions of new energy vehicles.
Smart Images

Figure CN118671634B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent safety detection of power batteries, and in particular to a data-driven lithium-ion battery fault diagnosis method and system. Background Art
[0002] As the world energy crisis intensifies, the development and use of new energy is the only way to achieve the dual carbon goals. Against this background, the new energy vehicle industry has developed rapidly. For new energy vehicles, power batteries provide power for them and are also related to the safety of the vehicle. Lithium batteries are widely used in new energy vehicles due to their high energy density and long service life. In recent years, car accidents caused by lithium-ion battery failures have occurred frequently, which has attracted widespread attention from the society.
[0003] Due to the complex internal mechanism of lithium-ion batteries and the changeable operating conditions of electric vehicles, the probability of battery failure is greatly increased, which seriously endangers the safety of people's lives and property. Lithium-ion battery failures mainly include sudden failures and progressive failures. Sudden failures occur in a short time and are extremely harmful. They are prone to thermal runaway, causing battery fire or explosion. The main causes of sudden failures are internal short circuits, external short circuits, excessive temperature, excessive pressure, overcharge and over-discharge, etc. Progressive failures are mostly manifested as high or low voltage and large pressure difference. The main reasons are battery aging, capacity attenuation, increased internal resistance, etc. The early manifestations of progressive failures are not obvious, but allowing them to develop will inevitably lead to greater accidents.
[0004] The failure of lithium-ion batteries usually manifests itself as changes in internal parameters. A data-driven approach can be adopted to collect battery parameters through sensors, extract relevant indicators, and establish a machine learning model for fault diagnosis. However, the existing fault diagnosis methods still have defects such as insufficient sensitivity to fault characteristics, poor real-time performance, high misjudgment rate, and inability to accurately determine the fault type in a timely manner. Therefore, a data-driven lithium-ion battery fault diagnosis method is proposed to solve the above problems. Summary of the invention
[0005] In view of the deficiencies in the prior art, the present invention provides a data-driven lithium-ion battery fault diagnosis method and system, which has the advantages of accurately and efficiently detecting battery faults and identifying fault types, and solves the problems of insufficient sensitivity to fault characteristics, poor real-time performance, and high misjudgment rate in existing fault diagnosis methods.
[0006] In order to achieve the above-mentioned purpose of accurately and efficiently detecting battery faults and identifying fault types, the present invention provides the following technical solutions: a data-driven lithium-ion battery fault diagnosis method, comprising the following steps:
[0007] S1: Collect voltage data during the actual operation of electric vehicles;
[0008] S2: preprocess the collected voltage data into the original time series X;
[0009] S3: extracting the first multi-scale fuzzy distribution entropy MFDE from the original time series X, and using the first multi-scale fuzzy distribution entropy MFDE as a test sample;
[0010] S4: construct a weighted least squares support vector machine model, and perform offline training on the weighted least squares support vector machine model to obtain a fault diagnosis discriminant function and a maximum model error, input the test sample into the weighted least squares support vector machine model, use the fault diagnosis discriminant function to obtain the fault diagnosis result, and calculate the test sample error;
[0011] S5: Determine whether the test sample error exceeds the maximum error of the model. If so, proceed to step S4, update the weighted least squares support vector machine model parameters, feed the test sample and the updated parameters back to the weighted least squares support vector machine model for re-diagnosis, until the test sample error is less than the maximum error of the model, proceed to step S6, and output the diagnosis result;
[0012] S6: If the diagnosis result is normal, the weighted least squares support vector machine model has no output, waits for the voltage data collection at the next moment, and repeats steps S1-S5. If the diagnosis result is a sudden fault, the weighted least squares support vector machine model output is 1; if the diagnosis result is a progressive fault, the weighted least squares support vector machine model output is 2.
[0013] Preferably, the specific steps of step S1 are:
[0014] Sensors are installed on the battery pack of the electric vehicle to form a sensor matrix, and the data sampling time is 5s.
[0015] Preferably, the specific steps of step S2 are:
[0016] S2.1: Clean the collected voltage data and remove the voltage data that exceeds a reasonable range;
[0017] S2.2: Normalize the voltage data after cleaning, where the specific expression of normalization is:
[0018] ;
[0019] Among them, x i is the original voltage data, x max is the maximum value of the original voltage data, x min is the minimum value of the original voltage data, x new (i) is the voltage data after normalization of the original data, The value range of is 1-N;
[0020] S2.3: Set all x new (i) Compose the original time series X.
[0021] Preferably, the specific steps of step S3 are:
[0022] S3.1: For the original time series X of length N, the scale factor τ is used for coarse-grained processing. The coarse-grained processing method adopts overlapping method. Each coarse-grained series The expression is:
[0023] ;
[0024] Where t is the summing factor;
[0025] S3.2: The length of all coarse-grained sequences is Multi-scale time series , reconstruct the phase space of the multi-scale time series F into T-m+1 m-dimensional reconstruction vectors, and calculate the average value y of the reconstruction vectors 0 (l);
[0026] ;
[0027] Where K is the summing factor;
[0028] S3.3: By reconstructing the vector and y 0 (l) Calculate T-m+1 m-dimensional state vectors Q(l);
[0029] ;
[0030] Where m is the embedding dimension;
[0031] S3.4: Define the distance matrix , calculate the Chebyshev distance between any two different state vectors Q(a) and Q(b) ;
[0032] ;
[0033] Where λ is the distance factor, a and b = 1, 2, ..., T-m+1;
[0034] S3.5: Define the similarity matrix , introduce a one-dimensional Gaussian function as the fuzzy membership function of the distance matrix, and calculate the similarity s between Q(a) and Q(b) ab :
[0035] ;
[0036] Where z is the exponential function order, r is the similarity tolerance, a and b = 1, 2, ..., T-m+1;
[0037] S3.6: Remove the elements on the diagonal of the similarity matrix S to reconstruct the similarity matrix, and use the histogram to perform frequency analysis on the reconstructed matrix to obtain the frequency distribution G h , calculate the empirical density function P h , calculate the empirical density function P h The expression is:
[0038] ;
[0039] S3.7: Calculation The fuzzy distribution entropy , the expression is:
[0040] ;
[0041] S3.8: The fuzzy distribution entropy FDE is the first multi-scale fuzzy distribution entropy MFDE of the original time series X;
[0042] ;
[0043] S3.9: Use the first multiscale fuzzy distribution entropy MFDE as a test sample.
[0044] Preferably, the specific steps of step S4 are:
[0045] S4.1: Construct a weighted least squares support vector machine model, map the samples to be tested to a high-dimensional space using a Gaussian kernel function, construct a classification hyperplane (ω, μ) in the high-dimensional space, and obtain the optimization objective function and constraint function;
[0046] Among them, ω is the hyperplane normal vector, μ is the intercept;
[0047] S4.2: Introduce fitting error to each sample to be tested , give different weights to the samples to be tested according to the size of the fitting error , the reconstruction target optimizes the objective function, and the weight reflects the contribution of the sample in the model construction process, which reduces the impact of outlier samples on the model and improves the robustness of the model;
[0048] S4.3: The improved crested porcupine optimization algorithm based on cosine variable adaptive weight factor is used to optimize the hyperparameters in the weighted least squares support vector machine model. The hyperparameters include the penalty factor C of the optimization objective function and the Gaussian kernel function parameter σ. The adaptive weight calculation formula of the improved crested porcupine optimization algorithm is:
[0049] ;
[0050] Among them, α is the adaptive inertia weight factor in the algorithm iteration process, γ is the current iteration number, and Max_iter is the maximum iteration number;
[0051] S4.4: Solve the weighted least squares support vector machine model, construct the Lagrangian function, find the partial derivative of its parameters and set the partial derivative value to 0, and obtain a simplified linear equation system. Solve the equation system to obtain the hyperplane intercept μ and Lagrangian parameters, and obtain the classification formula of the weighted least squares support vector machine model;
[0052] S4.5: Collect historical operating voltage information of normal and faulty vehicles, calculate the second multi-scale fuzzy distribution entropy MFDE using the method in step S3, store it as a feature data set, use the feature data set to train a weighted least squares support vector machine model, select the root mean square error as the error function of the weighted least squares support vector machine model, obtain the fault diagnosis discriminant function of the battery normal state and sudden fault, normal state and progressive fault, and obtain the maximum error of the model;
[0053] S4.6: Input the test sample into the weighted least squares support vector machine model for diagnosis to obtain the fault diagnosis test sample result and the test sample error.
[0054] Preferably, a data-driven lithium-ion battery fault diagnosis system comprises a signal acquisition module, a data processing module, a feature extraction module, a fault diagnosis module, a feedback optimization module and a diagnosis result display module;
[0055] The signal acquisition module is used to collect voltage data during the actual operation of the electric vehicle;
[0056] The data processing module is used to preprocess the collected data into the original time series X;
[0057] The feature extraction module is used to extract the first multi-scale fuzzy distribution entropy MFDE from the original time series X, and use the first multi-scale fuzzy distribution entropy MFDE as a test sample;
[0058] The fault diagnosis module is used to diagnose whether the test sample is faulty and identify the battery fault type;
[0059] The feedback optimization module is used to improve the accuracy of the weighted least squares support vector machine model and reduce the misjudgment rate;
[0060] The display module is used to display the diagnosis results.
[0061] Compared with the prior art, the present invention provides a data-driven lithium-ion battery fault diagnosis method, which has the following beneficial effects:
[0062] 1. This data-driven lithium-ion battery fault diagnosis method introduces fuzzy membership function into the distribution entropy, performs coarse-grained processing on the original time series, and extracts multi-scale fuzzy distribution entropy as fault characteristics. It can accurately reflect the voltage fluctuation characteristics of normal state, sudden fault and progressive fault, and adapt to various operating conditions of new energy vehicles.
[0063] 2. This data-driven lithium-ion battery fault diagnosis method adopts a weighted least squares support vector machine model in the fault diagnosis step, and uses an improved crown porcupine optimization algorithm based on cosine variable adaptive weight factors to optimize the penalty factor and kernel function parameters. The adaptive weight factor can be dynamically adjusted with the number of iterations, which can improve the algorithm accuracy and convergence speed, and has better real-time performance and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0064] Figure 1 A schematic diagram of a data-driven lithium-ion battery fault diagnosis method proposed in the present invention;
[0065] Figure 2 It is a multi-scale fuzzy distribution entropy curve diagram of a normal battery and a sudden fault battery of the present invention;
[0066] Figure 3 It is a multi-scale fuzzy distribution entropy curve diagram of a normal battery and a progressively faulty battery of the present invention. DETAILED DESCRIPTION
[0067] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0068] See also Figure 1-3 , a data-driven lithium-ion battery fault diagnosis method, comprising the following steps:
[0069] S1: Collect voltage data during the actual operation of electric vehicles;
[0070] S2: preprocess the collected voltage data into the original time series X;
[0071] S3: extracting the first multi-scale fuzzy distribution entropy MFDE from the original time series X, and using the first multi-scale fuzzy distribution entropy MFDE as a test sample;
[0072] S4: construct a weighted least squares support vector machine model, and perform offline training on the weighted least squares support vector machine model to obtain a fault diagnosis discriminant function and a maximum model error, input the test sample into the weighted least squares support vector machine model, use the fault diagnosis discriminant function to obtain the fault diagnosis result, and calculate the test sample error;
[0073] S5: Determine whether the test sample error exceeds the maximum error of the model. If so, proceed to step S4, update the weighted least squares support vector machine model parameters, feed the test sample and the updated parameters back to the weighted least squares support vector machine model for re-diagnosis, until the test sample error is less than the maximum error of the model, proceed to step S6, and output the diagnosis result;
[0074] S6: If the diagnosis result is normal, the weighted least squares support vector machine model has no output, waits for the voltage data collection at the next moment, and repeats steps S1-S5. If the diagnosis result is a sudden fault, the weighted least squares support vector machine model output is 1; if the diagnosis result is a progressive fault, the weighted least squares support vector machine model output is 2.
[0075] The specific steps of step S1 are:
[0076] Sensors are installed on the battery pack of the electric vehicle to form a sensor matrix, and the data sampling time is 5s.
[0077] The specific steps of step S2 are:
[0078] S2.1: Clean the collected voltage data and remove the voltage data that exceeds a reasonable range;
[0079] S2.2: Normalize the voltage data after cleaning, where the specific expression of normalization is:
[0080] ;
[0081] Among them, x i is the original voltage data, x max is the maximum value of the original voltage data, x mi is the minimum value of the original voltage data, x new (i) is the voltage data after normalization of the original data, The value range of is 1-N;
[0082] S2.3: Set all x new (i) Compose the original time series X.
[0083] The specific steps of step S3 are:
[0084] S3.1: For the original time series X of length N, the scale factor τ is used for coarse-grained processing. The coarse-grained processing method adopts overlapping method. Each coarse-grained series The expression is:
[0085] ;
[0086] Where t is the summing factor;
[0087] The complexity of voltage signals can be analyzed at multiple time scales to solve the problem of single voltage signal scale;
[0088] S3.2: The length of all coarse-grained sequences is Multi-scale time series Reconstruct the phase space of the multi-scale time series F into T-m+1 m-dimensional reconstruction vectors, and calculate the average value y of the reconstruction vectors 0 (l);
[0089] ;
[0090] Where K is the summing factor;
[0091] S3.3: By reconstructing the vector and y 0 (l) Calculate T-m+1 m-dimensional state vectors Q(l);
[0092] ;
[0093] Where m is the embedding dimension;
[0094] S3.4: Define the distance matrix ; Calculate the Chebyshev distance between any two different state vectors Q(a) and Q(b) ;
[0095] ;
[0096] Where λ is the distance factor, a and b = 1, 2, ..., T-m+1;
[0097] S3.5: Define the similarity matrix , introducing a one-dimensional Gaussian function as the fuzzy membership function of the distance matrix can more clearly reflect the membership relationship of each element in the distance matrix D, better express the fluctuation characteristics and nonlinear characteristics of the voltage signal, and calculate the similarity s between Q (a) and Q (b) ab :
[0098] ;
[0099] Where z is the exponential function order, r is the similarity tolerance, a and b = 1, 2, ..., T-m+1;
[0100] S3.6: To eliminate the influence of self-matching of matrix elements, remove the elements on the diagonal of the similarity matrix S to reconstruct the similarity matrix. Use the histogram to perform frequency analysis on the reconstructed matrix and obtain the frequency distribution G h , calculate the empirical density function P h , calculate the empirical density function P h The expression is:
[0101] ;
[0102] S3.7: Calculation The fuzzy distribution entropy ;
[0103] ;
[0104] S3.8: The fuzzy distribution entropy FDE is the first multi-scale fuzzy distribution entropy MFDE of the original time series X;
[0105] ;
[0106] S3.9: Use the first multiscale fuzzy distribution entropy MFDE as a test sample.
[0107] The specific steps of step S4 are:
[0108] S4.1: Construct a weighted least squares support vector machine model, map the samples to be tested to a high-dimensional space using a Gaussian kernel function, construct a classification hyperplane (ω, μ) in the high-dimensional space, and obtain the optimization objective function and constraint function;
[0109] Among them, ω is the hyperplane normal vector, μ is the intercept;
[0110] S4.2: Introduce fitting error to each sample to be tested , give different weights to the samples to be tested according to the size of the fitting error ,The reconstruction target optimizes the objective function, and the weight reflects the contribution of the sample in the model construction process, which reduces the impact of outlier samples on the model and improves the robustness of the model;
[0111] S4.3: The improved crested porcupine optimization algorithm based on cosine variable adaptive weight factor is used to optimize the hyperparameters in the weighted least squares support vector machine model. The hyperparameters include the penalty factor C of the optimization objective function and the Gaussian kernel function parameter σ. The adaptive weight calculation formula of the improved crested porcupine optimization algorithm is:
[0112] ;
[0113] Among them, α is the adaptive inertia weight factor in the algorithm iteration process, γ is the current iteration number, and Max_iter is the maximum iteration number;
[0114] S4.4: Solve the weighted least squares support vector machine model, construct the Lagrangian function, find the partial derivative of its parameters and set the partial derivative value to 0, and obtain a simplified linear equation system. Solve the equation system to obtain the hyperplane intercept μ and Lagrangian parameters, and obtain the classification formula of the weighted least squares support vector machine model;
[0115] S4.5: Collect historical operating voltage information of normal and faulty vehicles, calculate the second multi-scale fuzzy distribution entropy MFDE using the method in step S3, store it as a feature data set, use the feature data set to train a weighted least squares support vector machine model, select the root mean square error as the error function of the weighted least squares support vector machine model, obtain the fault diagnosis discriminant function of the battery normal state and sudden fault, normal state and progressive fault, and obtain the maximum error of the model;
[0116] S4.6: Input the test sample into the weighted least squares support vector machine model for diagnosis to obtain the fault diagnosis test sample result and the test sample error.
[0117] A data-driven lithium-ion battery fault diagnosis system, the system comprising a signal acquisition module, a data processing module, a feature extraction module, a fault diagnosis module, a feedback optimization module and a diagnosis result display module;
[0118] The signal acquisition module is used to collect voltage data during the actual operation of the electric vehicle;
[0119] The data processing module is used to preprocess the collected data into the original time series X;
[0120] The feature extraction module is used to extract the first multi-scale fuzzy distribution entropy MFDE from the original time series X, and use the first multi-scale fuzzy distribution entropy MFDE as a test sample;
[0121] The fault diagnosis module is used to diagnose whether the test sample is faulty and identify the battery fault type;
[0122] The feedback optimization module is used to improve the accuracy of the weighted least squares support vector machine model and reduce the misjudgment rate;
[0123] The display module is used to display the diagnosis results.
[0124] This invention, by introducing fuzzy membership function into distribution entropy, performs coarse-grained processing on the original time series, extracts multi-scale fuzzy distribution entropy as fault characteristics, can accurately reflect the voltage fluctuation characteristics of normal state, sudden fault and progressive fault, and adapt to various operating conditions of new energy vehicles;
[0125] In addition, by adopting the weighted least squares support vector machine model in the fault diagnosis step, the penalty factor and kernel function parameters are optimized using the improved crested porcupine optimization algorithm based on cosine variable adaptive weight factor. The adaptive weight factor can be dynamically adjusted with the change of the number of iterations, which can improve the algorithm accuracy and convergence speed, and has better real-time performance and accuracy.
[0126] It should be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.
[0127] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A data-driven lithium-ion battery fault diagnosis method, characterized in that: The following steps are involved: S1: Collect voltage data during the actual operation of electric vehicles; S2: preprocess the collected voltage data into the original time series X; S3: extracting the first multi-scale fuzzy distribution entropy MFDE from the original time series X, and using the first multi-scale fuzzy distribution entropy MFDE as a test sample; S4: construct a weighted least squares support vector machine model, and perform offline training on the weighted least squares support vector machine model to obtain a fault diagnosis discriminant function and a maximum model error, input the test sample into the weighted least squares support vector machine model, use the fault diagnosis discriminant function to obtain the fault diagnosis result, and calculate the test sample error; S5: Determine whether the test sample error exceeds the maximum error of the model. If so, proceed to step S4, update the weighted least squares support vector machine model parameters, feed the test sample and the updated parameters back to the weighted least squares support vector machine model for re-diagnosis, until the test sample error is less than the maximum error of the model, proceed to step S6, and output the diagnosis result; S6: If the diagnosis result is normal, the weighted least squares support vector machine model has no output, waits for the voltage data collection at the next moment, and repeats steps S1-S5. If the diagnosis result is a sudden fault, the weighted least squares support vector machine model output is 1; if the diagnosis result is a progressive fault, the weighted least squares support vector machine model output is 2; The specific steps of step S3 are: S3.1: For the original time series X of length N, the scale factor τ is used for coarse-grained processing. The coarse-grained processing method adopts overlapping method. Each coarse-grained sequence y τ (j) The expression is: S3.2: The length of all coarse-grained sequences is Multi-scale time series , reconstruct the phase space of the multi-scale time series F into T-m+1 m-dimensional reconstruction vectors, and calculate the average value y0(l) of the reconstruction vectors; Where K is the summing factor; S3.3: Calculate T-m+1 m-dimensional state vectors Q(l) by reconstructing the vector and y0(l); Where m is the embedding dimension; S3.4: Define the distance matrix , calculate the Chebyshev distance between any two different state vectors Q(a) and Q(b) ; Where λ is the distance factor, a and b = 1, 2, ..., T-m+1; S3.5: Define the similarity matrix , introduce a one-dimensional Gaussian function as the fuzzy membership function of the distance matrix, and calculate the similarity s between Q(a) and Q(b) ab : Where z is the exponential function order, r is the similarity tolerance, a and b = 1, 2, ..., T-m+1; S3.6: Remove the elements on the diagonal of the similarity matrix S to reconstruct the similarity matrix, and use the histogram to perform frequency analysis on the reconstructed matrix to obtain the frequency distribution G h , calculate the empirical density function P h , calculate the empirical density function P h The expression is: S3.7: Calculate y τ (j) Fuzzy distribution entropy , the expression is: S3.8:y τ The fuzzy distribution entropy FDE of (j) is the first multi-scale fuzzy distribution entropy MFDE of the original time series X; S3.9: Use the first multiscale fuzzy distribution entropy MFDE as a test sample.
2. A data-driven lithium-ion battery fault diagnosis method according to claim 1, characterized in that: The specific steps of step S1 are: Sensors are installed on the battery pack of the electric vehicle to form a sensor matrix, and the data sampling time is 5s.
3. The data-driven lithium-ion battery fault diagnosis method according to claim 1, characterized in that: The specific steps of step S2 are: S2.1: Clean the collected voltage data and remove the voltage data that exceeds a reasonable range; S2.2: Normalize the voltage data after cleaning, where the specific expression of normalization is: Among them, x i is the original voltage data, x max is the maximum value of the original voltage data, x min is the minimum value of the original voltage data, x new (i) is the voltage data after normalization of the original data, and the value range of i is 1-N; S2.3: Set all x new (i) Compose the original time series X.
4. A data-driven lithium-ion battery fault diagnosis method according to claim 1, characterized in that ,The specific steps of step S4 are: S4.1: Construct a weighted least squares support vector machine model, map the samples to be tested to a high-dimensional space using a Gaussian kernel function, construct a classification hyperplane (ω, μ) in the high-dimensional space, and obtain the optimization objective function and constraint function; Among them, ω is the hyperplane normal vector, μ is the intercept; S4.2: Introduce fitting error to each sample to be tested , give different weights to the samples to be tested according to the size of the fitting error , reconstruct the target optimization objective function; S4.3: The improved crown porcupine optimization algorithm based on cosine variable adaptive weight factor is used to optimize the hyperparameters in the weighted least squares support vector machine model. The hyperparameters include the penalty factor C of the optimization objective function and the Gaussian kernel function parameter σ. The adaptive weight calculation formula of the improved crown porcupine optimization algorithm is: Among them, α is the adaptive inertia weight factor in the algorithm iteration process, γ is the current iteration number, and Max_iter is the maximum iteration number; S4.4: Solve the weighted least squares support vector machine model, construct the Lagrangian function, find the partial derivative of its parameters and set the partial derivative value to 0, and obtain a simplified linear equation system. Solve the equation system to obtain the hyperplane intercept μ and Lagrangian parameters, and obtain the classification formula of the weighted least squares support vector machine model; S4.5: Collect historical operating voltage information of normal and faulty vehicles, calculate the second multi-scale fuzzy distribution entropy MFDE using the method in step S3, store it as a feature data set, use the feature data set to train a weighted least squares support vector machine model, select the root mean square error as the error function of the weighted least squares support vector machine model, obtain the fault diagnosis discriminant function of the battery normal state and sudden fault, normal state and progressive fault, and obtain the maximum error of the model; S4.6: Input the test sample into the weighted least squares support vector machine model for diagnosis to obtain the fault diagnosis test sample result and the test sample error.
5. A data-driven lithium-ion battery fault diagnosis system based on any one of the methods of claims 1 to 4, characterized in that: The system includes a signal acquisition module, a data processing module, a feature extraction module, a fault diagnosis module, a feedback optimization module and a diagnosis result display module; The signal acquisition module is used to collect voltage data during the actual operation of the electric vehicle; The data processing module is used to preprocess the collected data into the original time series X; The feature extraction module is used to extract the first multi-scale fuzzy distribution entropy MFDE from the original time series X, and use the first multi-scale fuzzy distribution entropy MFDE as a test sample; The fault diagnosis module is used to diagnose whether the test sample is faulty and identify the battery fault type; The feedback optimization module is used to improve the accuracy of the weighted least squares support vector machine model and reduce the misjudgment rate; The display module is used to display the diagnosis results.
