An electric vehicle power battery fault diagnosis method, system, device and medium
By constructing multiple fault diagnosis models based on different neural networks, and combining voting factors, decision weights, and decision thresholds, a multi-level decision fusion algorithm was adopted to solve the problem of low accuracy of single neural network models in the fault diagnosis of electric vehicle power batteries, thus achieving higher diagnostic accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA JILIANG UNIV
- Filing Date
- 2023-02-16
- Publication Date
- 2026-06-26
AI Technical Summary
In existing technologies, single neural network models do not perform well in classifying electric vehicle power battery faults, and traditional model fusion methods such as voting methods may ignore the effective information of a few models, resulting in low diagnostic accuracy.
Multiple fault diagnosis models based on different neural networks are used. By acquiring battery parameters, calculating voting factors, decision weights and decision thresholds, and combining a multi-level decision fusion algorithm, the classification information of each model is comprehensively utilized to select the final fault diagnosis result.
It improves the accuracy of fault diagnosis for electric vehicle power batteries by comprehensively utilizing information from various models through a multi-level decision fusion algorithm, thus achieving more reliable fault diagnosis.
Smart Images
Figure CN116819328B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of battery fault diagnosis, and in particular to a method, system, device and medium for diagnosing faults in electric vehicle power batteries. Background Technology
[0002] Driven by both environmental protection and energy conservation, the low-carbon economy has extended to the transportation sector. Currently, the new energy vehicle industry is developing rapidly. From January to September 2022, production and sales of new energy vehicles reached 4.717 million and 4.567 million units respectively, representing year-on-year increases of 1.2 times and 1.1 times, with a market share of 23.5%. Electric vehicles, with their clean energy and low energy consumption, have become an important new type of green and environmentally friendly transportation tool.
[0003] The power battery is the sole power source for electric vehicles, and its safety and reliability are closely related to the safety and reliability of the entire vehicle system. Therefore, it is necessary to monitor various battery parameters in real time and diagnose and handle battery faults promptly. Power batteries have an electrochemical structure, and serious faults can lead to permanent damage to the battery pack. Therefore, research on power battery fault diagnosis has significant application value and practical importance.
[0004] Currently, research on fault diagnosis in power batteries mainly focuses on expert diagnosis and neural networks. Some literature uses expert systems to build an overall framework and employs fuzzy mathematics and neural networks to achieve battery fault diagnosis. Many other studies use BP neural networks, RBF neural networks, probabilistic neural networks, fuzzy neural networks, and combinations of neural networks with other methods or optimization algorithms for battery fault diagnosis. Power batteries are real-time changing nonlinear systems, and their performance is affected by changes in various parameters. Neural networks can process large-scale data in parallel and are an important tool for handling multi-input, multi-output nonlinear real-variable systems. However, each neural network has its advantages and disadvantages; a single neural network usually only has good classification results for a few types. Therefore, the classification results obtained by fusing multiple neural network models through fusion algorithms are more reliable and also achieve information complementarity between different neural network models.
[0005] Model fusion methods mainly include the maximum value method, product method, summation method, mean method, and voting method. The first four fusion methods all require the output of the classification model to be in the form of probability values, and are not suitable for classification models whose output directly represents the category to which the sample belongs. The arbitration principle of the voting method is "majority rule," which may ignore the effective classification information of the minority models if it is strictly a matter of majority rule. Therefore, how to improve the fault diagnosis accuracy of electric vehicle power batteries based on model fusion methods has become an urgent problem to be solved. Summary of the Invention
[0006] Based on this, embodiments of the present invention provide a method, system, device and medium for diagnosing faults in electric vehicle power batteries, so as to improve the accuracy of fault diagnosis of electric vehicle power batteries.
[0007] To achieve the above objectives, the present invention provides the following solution:
[0008] A method for diagnosing faults in electric vehicle power batteries, comprising:
[0009] Obtain the battery parameters of the target battery; the battery parameters include: battery current, individual cell voltage, battery state of charge, total battery voltage, battery temperature, and the trend of total battery voltage change;
[0010] The battery parameters of the target battery are input into multiple fault diagnosis models to obtain multiple fault diagnosis results for the target battery; the multiple fault diagnosis models are constructed based on different neural networks; the fault diagnosis results include: fault type;
[0011] Based on multiple fault diagnosis results, a voting factor for the fault type of the target battery output by each fault diagnosis model is determined; the voting factor for each fault type is obtained by testing multiple fault diagnosis models using a test set; the test set includes test samples and the actual fault type corresponding to each test sample;
[0012] Determine the decision weights for each fault diagnosis model; the decision weights are obtained by testing multiple fault diagnosis models using the test set.
[0013] The decision threshold is determined based on the voting factors for various fault types obtained by testing each fault diagnosis model using the test set.
[0014] Based on the voting factors of the target battery's fault type output by each fault diagnosis model, the decision weights of each fault diagnosis model, and the decision threshold, the final fault diagnosis result of the target battery is selected from multiple fault diagnosis results.
[0015] Optionally, based on the voting factors for the fault type of the target battery output by each fault diagnosis model, the decision weights of each fault diagnosis model, and the decision threshold, the final fault diagnosis result of the target battery is selected from multiple fault diagnosis results, specifically including:
[0016] The maximum value among the voting factors for the fault type of the target battery output by each fault diagnosis model is determined as the optimal voting factor, and the fault diagnosis model corresponding to the optimal voting factor is determined as the preliminary optimal diagnosis model.
[0017] Determine whether the difference between the preliminary optimal diagnostic model and the remaining diagnostic models is greater than the decision threshold to obtain a first judgment result; the remaining diagnostic models are models other than the preliminary optimal diagnostic model among the multiple fault diagnosis models.
[0018] If the first judgment result is yes, then the fault diagnosis result output by the preliminary optimal diagnosis model shall be taken as the final fault diagnosis result of the target battery.
[0019] If the first judgment result is negative, then it is determined whether all of the multiple fault diagnosis results are different, and a second judgment result is obtained.
[0020] If the second judgment result is negative, then the principle of majority rule is adopted to determine the final fault diagnosis result of the target battery from the multiple fault diagnosis results;
[0021] If the second judgment result is yes, then the decision weights of the preliminary optimal diagnostic model and the remaining diagnostic models are compared, and the fault diagnosis result output by the fault diagnosis model corresponding to the largest decision weight is taken as the final fault diagnosis result of the target battery.
[0022] Optionally, based on multiple fault diagnosis results, a voting factor is determined for the fault type of the target battery output by each fault diagnosis model, specifically including:
[0023] Obtain the test set;
[0024] The test set is input into each fault diagnosis model, and each fault diagnosis model outputs the predicted fault type for each test sample.
[0025] For any fault diagnosis model, the confidence level of the fault diagnosis model output for each fault type is calculated based on the actual fault type and the predicted fault type of each test sample.
[0026] For any fault type, calculate the average confidence level of the fault type output by multiple fault diagnosis models based on the confidence levels of various fault types output by all fault diagnosis models.
[0027] For any fault diagnosis model, the voting factor for the fault diagnosis model outputting a certain fault type is calculated based on the prediction precision of the fault diagnosis model for a certain fault type, the prediction recall of the fault diagnosis model for a certain fault type, and the average confidence.
[0028] For any fault diagnosis model, select the voting factor that matches the fault type of the corresponding fault diagnosis result from the voting factors of each fault type output by the fault diagnosis model, and use it as the voting factor of the fault type of the target battery output by the fault diagnosis model.
[0029] Optionally, the decision weights for each fault diagnosis model are determined, specifically including:
[0030] For any fault diagnosis model, based on the actual fault type and predicted fault type of each test sample output by the fault diagnosis model, determine the number of correctly classified test samples and the number of incorrectly classified test samples.
[0031] For any fault diagnosis model, the ratio of the number of correctly classified test samples to the total number of test samples is determined as the decision weight of the fault diagnosis model.
[0032] Optionally, the formula for calculating the decision threshold is:
[0033] ε=λ*max(e1,e2,...e m )
[0034] Where ε is the decision threshold; λ is the decision threshold coefficient; λ∈[0,1]; e1 is the set of voting factors for various fault types output by the first fault diagnosis model; e2 is the set of voting factors for various fault types output by the second fault diagnosis model; e m Output the set of voting factors for various fault types for the m-th fault diagnosis model; m represents the number of fault diagnosis models.
[0035] Optionally, the fault diagnosis model is three;
[0036] The first fault diagnosis model is based on a convolutional neural network; the second fault diagnosis model is based on a backpropagation neural network; and the third fault diagnosis model is based on an RBF neural network.
[0037] The present invention also provides a fault diagnosis system for electric vehicle power batteries, comprising:
[0038] The battery parameter acquisition module is used to acquire the battery parameters of the target battery; the battery parameters include: battery current, individual cell voltage, battery state of charge, total battery voltage, battery temperature, and the trend of total battery voltage change.
[0039] The fault diagnosis module is used to input the battery parameters of the target battery into multiple fault diagnosis models to obtain multiple fault diagnosis results of the target battery; the multiple fault diagnosis models are constructed based on different neural networks; the fault diagnosis results include: fault type;
[0040] The voting factor determination module is used to determine the voting factor for the fault type of the target battery output by each fault diagnosis model based on multiple fault diagnosis results. The voting factor for each fault type is obtained by testing multiple fault diagnosis models using a test set. The test set includes test samples and the actual fault type corresponding to each test sample.
[0041] The decision weight determination module is used to determine the decision weight of each fault diagnosis model; the decision weight is obtained by testing multiple fault diagnosis models using the test set;
[0042] The decision threshold determination module is used to determine the decision threshold based on the voting factors of various fault types obtained by testing each fault diagnosis model using the test set;
[0043] The diagnostic result fusion module is used to select the final fault diagnosis result of the target battery from multiple fault diagnosis results based on the voting factor of the fault type of the target battery output by each fault diagnosis model, the decision weight of each fault diagnosis model and the decision threshold.
[0044] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the above-described electric vehicle power battery fault diagnosis method.
[0045] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for diagnosing faults in electric vehicle power batteries.
[0046] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0047] This invention proposes a method, system, device, and medium for fault diagnosis of electric vehicle power batteries. The method involves acquiring battery parameters of the target battery; inputting these parameters into multiple fault diagnosis models to obtain multiple fault diagnosis results; determining voting factors for the fault types of the target battery output by each fault diagnosis model, decision weights for each model, and a decision threshold based on these results; and achieving multi-level decision fusion of multiple fault diagnosis results based on the voting factors, decision weights, and decision thresholds to obtain the final fault diagnosis result, thereby improving the accuracy of fault diagnosis for electric vehicle power batteries. Attached Figure Description
[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0049] Figure 1 A flowchart of a method for diagnosing electric vehicle power battery faults provided in an embodiment of the present invention;
[0050] Figure 2 This is a structural diagram of an electric vehicle power battery fault diagnosis system provided in an embodiment of the present invention. Detailed Implementation
[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0052] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0053] Example 1
[0054] See Figure 1 The electric vehicle power battery fault diagnosis method of this embodiment includes:
[0055] Step 101: Obtain the battery parameters of the target battery; the battery parameters include: battery current, individual cell voltage, battery state of charge, total battery voltage, battery temperature, and the trend of total battery voltage change.
[0056] Step 102: Input the battery parameters of the target battery into multiple fault diagnosis models to obtain multiple fault diagnosis results of the target battery; the multiple fault diagnosis models are constructed based on different neural networks.
[0057] The fault diagnosis results include: fault type. Fault types include: normal state and fault state; fault states include: overcharge, over-discharge, high temperature, overvoltage, and overcurrent.
[0058] Step 103: Based on the multiple fault diagnosis results, determine the voting factor for the fault type of the target battery output by each fault diagnosis model.
[0059] The voting factor for each fault type is obtained by testing multiple fault diagnosis models using a test set; the test set includes test samples and the actual fault types corresponding to each test sample.
[0060] Step 103 specifically includes:
[0061] 1) Obtain the test set.
[0062] 2) Input the test set into each fault diagnosis model, and each fault diagnosis model outputs the predicted fault type of each test sample.
[0063] 3) For any fault diagnosis model, based on the actual fault type and predicted fault type of each test sample, calculate the confidence level of the model's output for each fault type. Wherein, the confidence level β of the j-th fault diagnosis model outputting fault type i is... ji The calculation formula is:
[0064]
[0065] x i x represents the number of test samples for fault type i. ji This represents the number of test samples that are input into the j-th fault diagnosis model and classified as fault type i.
[0066] 4) For any fault type, based on the reliability of various fault types output by all fault diagnosis models, calculate the average reliability of the fault type output by multiple fault diagnosis models. The average reliability of the j-th fault diagnosis model outputting fault type i... The calculation formula is:
[0067]
[0068] Where m represents the number of fault diagnosis models.
[0069] 5) For any fault diagnosis model, calculate the voting factor for a certain fault type output by the fault diagnosis model based on the prediction precision of the fault diagnosis model for a certain fault type, the prediction recall of the fault diagnosis model for a certain fault type, and the average confidence.
[0070] The formulas for calculating prediction precision and prediction recall are as follows:
[0071]
[0072]
[0073] This represents the prediction accuracy of the j-th fault diagnosis model for fault type i. Prediction accuracy is used to measure the accuracy of the fault diagnosis model for a certain category of diagnosis results. Let j represent the prediction recall of the j-th fault diagnosis model for fault type i. Prediction recall is used to measure the coverage of the fault diagnosis model for a certain category of diagnosis results; j = 1, 2, ..., m; k represents the number of fault types. This represents the number of test samples in which the j-th fault diagnosis model diagnoses fault type i as fault type i; This represents the number of test samples in which the j-th fault diagnosis model diagnoses fault type i as fault type s; This represents the number of test samples in which the j-th fault diagnosis model diagnoses fault type s as fault type i.
[0074] Among them, the voting factor E of the j-th fault diagnosis model outputs the fault type i. ji The calculation formula is:
[0075]
[0076] 6) For any fault diagnosis model, select the voting factor that is consistent with the fault type of the corresponding fault diagnosis result from the voting factors of each fault type output by the fault diagnosis model, and use it as the voting factor of the fault type of the target battery output by the fault diagnosis model.
[0077] Step 104: Determine the decision weights of each fault diagnosis model; the decision weights are obtained by testing multiple fault diagnosis models using the test set.
[0078] Step 104 specifically includes:
[0079] For any fault diagnosis model, based on the actual fault type and predicted fault type of each test sample output by the fault diagnosis model, the number of correctly classified test samples and the number of misclassified test samples are determined. For any fault diagnosis model, the ratio of the number of correctly classified test samples to the total number of test samples is determined as the decision weight of the fault diagnosis model. The total number of test samples is the sum of the number of correctly classified test samples and the number of misclassified test samples.
[0080] Wherein, the decision weight A of the j-th fault diagnosis model j The calculation formula is:
[0081]
[0082] Step 105: Determine the decision threshold based on the voting factors for various fault types obtained by testing each fault diagnosis model using the test set. The formula for calculating the decision threshold is:
[0083] ε=λ*max(e1,e2,...e m (7)
[0084] Where ε is the decision threshold; λ is the decision threshold coefficient; λ∈[0,1]; e1 is the set of voting factors for various fault types output by the first fault diagnosis model; e2 is the set of voting factors for various fault types output by the second fault diagnosis model; e m Output the set of voting factors for various fault types for the m-th fault diagnosis model; m represents the number of fault diagnosis models.
[0085] Step 106: Based on the voting factors of the target battery's fault type output by each fault diagnosis model, the decision weights of each fault diagnosis model, and the decision threshold, select the final fault diagnosis result of the target battery from the multiple fault diagnosis results.
[0086] Step 106 specifically includes:
[0087] 1) The maximum value among the voting factors of the target battery's fault type output by each fault diagnosis model is determined as the optimal voting factor, and the fault diagnosis model corresponding to the optimal voting factor is determined as the preliminary optimal diagnosis model.
[0088] 2) Determine whether the difference between the preliminary optimal diagnostic model and the remaining diagnostic model is greater than the decision threshold to obtain a first judgment result; the remaining diagnostic model is a model other than the preliminary optimal diagnostic model among the multiple fault diagnostic models.
[0089] If the first judgment result is yes, then the fault diagnosis result output by the preliminary optimal diagnosis model is taken as the final fault diagnosis result of the target battery; if the first judgment result is no, then step 3 is executed.
[0090] 3) Determine whether the multiple fault diagnosis results are all different to obtain a second judgment result.
[0091] If the second judgment result is negative, then the principle of majority rule is adopted to determine the final fault diagnosis result of the target battery from the multiple fault diagnosis results.
[0092] If the second judgment result is yes, then the decision weights of the preliminary optimal diagnostic model and the remaining diagnostic models are compared, and the fault diagnosis result output by the fault diagnosis model corresponding to the largest decision weight is taken as the final fault diagnosis result of the target battery.
[0093] In one example, there are three fault diagnosis models; the first fault diagnosis model is built based on a convolutional neural network (CNN); the second fault diagnosis model is built based on a backpropagation neural network; and the third fault diagnosis model is built based on an RBF neural network.
[0094] The following section uses three fault diagnosis models as examples to explain in detail a specific process of the electric vehicle power battery fault diagnosis method in practical application.
[0095] Step 1: Prepare and process the dataset. The battery management system monitors various parameters of the power battery in real time, including voltage, current, temperature, capacity, and state of charge (SOC). Export these parameter information. Set the input parameters of the dataset as: battery current, individual cell voltage, battery SOC, total battery voltage, battery temperature, and the trend of total battery voltage change (represented by values in the range [-1, 1]). The main fault modes of the battery are overcharge, over-discharge, high temperature, overvoltage, and overcurrent. Battery faults are uncertain and multiple faults may occur simultaneously. The fault type and its severity can be determined by monitoring the battery parameters. In this embodiment, the fault type with the highest severity is taken as the final fault type. In actual diagnosis, to represent all states of the battery, the normal state of the battery is added. Therefore, the output information of the dataset is the various states of the battery (one normal state and the above five fault states). The above six battery states are represented by six 0 or 1 digits. A digit of 0 in the corresponding position of the battery state indicates that no fault has occurred, and a digit of 1 indicates that such a fault has occurred. The input parameters and output information correspond one-to-one.
[0096] Step 2: Divide the dataset into a training set and a test set in a 9:1 ratio. The training set samples are used to train and optimize the model, and the test set samples are used to verify the accuracy of the model.
[0097] Step 3: Construct base class models. In this embodiment, three types of neural networks are selected as base class models: convolutional neural network, backpropagation neural network, and RBF neural network. The number of neurons in the input layer of the neural network is 8, and the number of neurons in the output layer is 6. The three base class models are trained using the training set to continuously optimize the network structure. After training, the diagnostic results of each model are obtained using the test set.
[0098] Step 4: Multi-model fusion diagnosis. A multi-level decision fusion algorithm is used to fuse the diagnostic results of the three base model types, outputting the final diagnostic result. The implementation process of the multi-level decision fusion algorithm is as follows:
[0099] (1) First, calculate the confidence level of each fault type (including normal state) output by each base class model. The confidence level β of fault type i output by the j-th base class model (fault diagnosis model) is calculated. ji The above formula (1) can be used for calculation. Then, the average confidence level of the fault type i output by the m base class models is calculated according to formula (2). In this embodiment, m = 3.
[0100] (2) Calculate the voting factors of each fault type output by each base class model. The voting factors can be calculated according to formula (3), formula (4), and formula (5).
[0101] (3) During fault diagnosis, compare the voting factors of the output results of each base class model, conduct multi-level decision voting, assign voting weight to each base class model, and take the output result of the base class model with the larger voting weight as the final diagnosis result.
[0102] The process of multi-level decision-making voting is as follows:
[0103] (1) Set a decision threshold ε, which is used when there is no significant difference in the voting factors between the two base classifiers, and the result of the base classification model needs to be judged at the next level to determine the final diagnosis result. In this embodiment, m=3, so the corresponding decision threshold ε is determined by formula (8).
[0104] ε=λ*max(e1,e2,e3) (8)
[0105] λ is the decision threshold coefficient, which is taken as λ times the largest voting factor in the three base classification models, where λ∈[0,1].
[0106] (2) Find the base classification model 'a' with the largest voting factor. When the difference between the voting factors of model 'a' and the other base class models is greater than ε, assign a voting weight of 1 to base class model 'a', and assign a voting weight of 0 to the other base class models. Otherwise, proceed to the next level of decision-making—step (3). Specifically:
[0107]
[0108] V j E represents the voting weight of the j-th base classification model. j E1 represents the voting factor for the fault type of the target battery (i.e., the battery to be diagnosed in actual diagnosis) output by the j-th base classification model; E2 represents the voting factor for the fault type of the target battery output by the 1st base classification model; and E3 represents the voting factor for the fault type of the target battery output by the 2nd base classification model.
[0109] (3) When there are identical diagnostic results in the base class model, the majority rule is applied, and the majority category T is identified from the diagnostic results. Specifically:
[0110]
[0111] T j This represents the fault category diagnosed by the j-th base classification model. When the output results of the three base classification models are different, jump to the next level decision step (4).
[0112] (4) Let b represent the set of base class models whose difference in voting factors with base classification model a is less than or equal to ε. Compare the decision weights of model a with those of the models in set b. The decision weights can be calculated using formula (6). Specifically:
[0113]
[0114] A a For the decision weights of the base classification model a, A b Let be the decision weights corresponding to each model in set b.
[0115] (5) Based on the above multi-level decision-making, take the diagnostic result of the classification model with the largest voting weight.
[0116] Step 5: Obtain the accuracy of the multi-model fusion diagnostic method.
[0117] The electric vehicle power battery fault diagnosis method in this embodiment employs an improved voting method—a multi-level decision fusion algorithm. It calculates the voting factors of the output results of each base class model, sets a decision threshold, and performs multi-level decision voting based on the threshold. Each base class model is assigned a voting weight, and the output result of the base class model with the highest voting weight is taken as the final diagnosis result. This method uses a multi-level decision algorithm to fuse the diagnostic results of multiple classification models, comprehensively and effectively utilizing the classification information of each model to improve the accuracy of electric vehicle power battery fault diagnosis. This method can be applied to power battery management systems for accurate battery fault diagnosis and fault cause analysis and early warning, thereby improving the safety of power battery use.
[0118] Example 2
[0119] In order to implement the method corresponding to Embodiment 1 above and achieve the corresponding functions and technical effects, an electric vehicle power battery fault diagnosis system is provided below.
[0120] See Figure 2 The system includes:
[0121] The battery parameter acquisition module 201 is used to acquire the battery parameters of the target battery; the battery parameters include: battery current, battery cell voltage, battery state of charge, battery total voltage, battery temperature, and battery total voltage change trend.
[0122] The fault diagnosis module 202 is used to input the battery parameters of the target battery into multiple fault diagnosis models to obtain multiple fault diagnosis results of the target battery; the multiple fault diagnosis models are constructed based on different neural networks; the fault diagnosis results include: fault type.
[0123] The voting factor determination module 203 is used to determine the voting factor of the target battery output by each fault diagnosis model based on the multiple fault diagnosis results. The voting factor of each fault type is obtained by testing multiple fault diagnosis models using a test set. The test set includes test samples and the actual fault type corresponding to each test sample.
[0124] The decision weight determination module 204 is used to determine the decision weight of each fault diagnosis model; the decision weight is obtained by testing multiple fault diagnosis models using the test set.
[0125] The decision threshold determination module 205 is used to determine the decision threshold based on the voting factors of various fault types obtained by testing each fault diagnosis model using the test set.
[0126] The diagnostic result fusion module 206 is used to select the final fault diagnosis result of the target battery from multiple fault diagnosis results based on the voting factor of the fault type of the target battery output by each fault diagnosis model, the decision weight of each fault diagnosis model and the decision threshold.
[0127] Example 3
[0128] This embodiment provides an electronic device, including a memory and a processor. The memory is used to store computer programs, and the processor runs the computer programs to enable the electronic device to perform the electric vehicle power battery fault diagnosis method of Embodiment 1.
[0129] Alternatively, the aforementioned electronic device may be a server.
[0130] In addition, embodiments of the present invention also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the electric vehicle power battery fault diagnosis method of Embodiment 1.
[0131] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.
[0132] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A method for diagnosing faults in the power battery of an electric vehicle, characterized in that, include: Obtain the battery parameters of the target battery; The battery parameters include: battery current, individual cell voltage, battery state of charge, total battery voltage, battery temperature, and the trend of total battery voltage variation. The battery parameters of the target battery are input into multiple fault diagnosis models to obtain multiple fault diagnosis results for the target battery; the multiple fault diagnosis models are constructed based on different neural networks; the fault diagnosis results include: fault type; Based on multiple fault diagnosis results, a voting factor for the fault type of the target battery output by each fault diagnosis model is determined. The voting factor for each fault type is obtained by testing multiple fault diagnosis models using a test set. The test set includes test samples and the actual fault types corresponding to each test sample. For any fault diagnosis model, the voting factor for each fault type is calculated based on the prediction precision of the fault diagnosis model for each fault type, the prediction recall of the fault diagnosis model for each fault type, and the average confidence of multiple fault diagnosis models for each fault type. Determine the decision weights for each fault diagnosis model; the decision weights are obtained by testing multiple fault diagnosis models using the test set. The decision threshold is determined based on the voting factors for various fault types obtained by testing each fault diagnosis model using the test set. Based on the voting factors of the fault type of the target battery output by each fault diagnosis model, the decision weights of each fault diagnosis model and the decision threshold, multi-level decision fusion is performed on the fault diagnosis results output by multiple fault diagnosis models to select the final fault diagnosis result of the target battery. Determine the decision weights for each fault diagnosis model, specifically including: For any fault diagnosis model, based on the actual fault type and predicted fault type of each test sample output by the fault diagnosis model, determine the number of correctly classified test samples and the number of incorrectly classified test samples. For any fault diagnosis model, the ratio of the number of correctly classified test samples to the total number of test samples is determined as the decision weight of the fault diagnosis model. Among them, the Each fault diagnosis model outputs the fault type. Credibility The calculation formula is: ； Indicates the fault type The number of test samples, Indicates that the test set is input to the first... Each fault diagnosis model is classified into fault types. The number of test samples; No. Each fault diagnosis model outputs the fault type. Average credibility The calculation formula is: ； in, m This indicates the number of fault diagnosis models.
2. The method for diagnosing faults in electric vehicle power batteries according to claim 1, characterized in that, Based on the voting factors for the fault types of the target battery output by each fault diagnosis model, the decision weights of each fault diagnosis model, and the decision threshold, the final fault diagnosis result of the target battery is selected from multiple fault diagnosis results, specifically including: The maximum value among the voting factors for the fault type of the target battery output by each fault diagnosis model is determined as the optimal voting factor, and the fault diagnosis model corresponding to the optimal voting factor is determined as the preliminary optimal diagnosis model. Determine whether the difference between the voting factors of the preliminary optimal diagnostic model and the remaining diagnostic models is greater than the decision threshold to obtain a first judgment result; the remaining diagnostic models are models other than the preliminary optimal diagnostic model among the multiple fault diagnosis models. If the first judgment result is yes, then the fault diagnosis result output by the preliminary optimal diagnosis model shall be taken as the final fault diagnosis result of the target battery. If the first judgment result is negative, then it is determined whether all of the multiple fault diagnosis results are different, and a second judgment result is obtained. If the second judgment result is negative, then the principle of majority rule is adopted to determine the final fault diagnosis result of the target battery from the multiple fault diagnosis results; If the second judgment result is yes, then the decision weights of the preliminary optimal diagnostic model and the remaining diagnostic models are compared, and the fault diagnosis result output by the fault diagnosis model corresponding to the largest decision weight is taken as the final fault diagnosis result of the target battery.
3. The method for diagnosing faults in electric vehicle power batteries according to claim 1, characterized in that, Based on multiple fault diagnosis results, a voting factor is determined for the fault type of the target battery output by each fault diagnosis model, specifically including: Obtain the test set; The test set is input into each fault diagnosis model, and each fault diagnosis model outputs the predicted fault type for each test sample. For any fault diagnosis model, the confidence level of the fault diagnosis model output for each fault type is calculated based on the actual fault type and the predicted fault type of each test sample. For any fault type, calculate the average confidence level of the fault type output by multiple fault diagnosis models based on the confidence levels of various fault types output by all fault diagnosis models. For any fault diagnosis model, the voting factor for the fault diagnosis model outputting a certain fault type is calculated based on the prediction precision of the fault diagnosis model for a certain fault type, the prediction recall of the fault diagnosis model for a certain fault type, and the average confidence. For any fault diagnosis model, select the voting factor that matches the fault type of the corresponding fault diagnosis result from the voting factors of each fault type output by the fault diagnosis model, and use it as the voting factor of the fault type of the target battery output by the fault diagnosis model.
4. The method for diagnosing faults in electric vehicle power batteries according to claim 1, characterized in that, The formula for calculating the decision threshold is: in, This is the decision threshold; This is the decision threshold coefficient; ; Output a set of voting factors for various fault types for the first fault diagnosis model; Output a set of voting factors for various fault types for the second fault diagnosis model; Output a set of voting factors for various fault types for the m-th fault diagnosis model; m This indicates the number of fault diagnosis models.
5. The method for diagnosing faults in electric vehicle power batteries according to claim 1, characterized in that, There are three fault diagnosis models; The first fault diagnosis model is based on a convolutional neural network; the second fault diagnosis model is based on a backpropagation neural network; and the third fault diagnosis model is based on an RBF neural network.
6. A fault diagnosis system for electric vehicle power batteries, characterized in that, include: The battery parameter acquisition module is used to acquire the battery parameters of the target battery. The battery parameters include: battery current, individual cell voltage, battery state of charge, total battery voltage, battery temperature, and the trend of total battery voltage variation. The fault diagnosis module is used to input the battery parameters of the target battery into multiple fault diagnosis models to obtain multiple fault diagnosis results of the target battery; the multiple fault diagnosis models are constructed based on different neural networks; the fault diagnosis results include: fault type; The voting factor determination module is used to determine the voting factor of the fault type of the target battery output by each fault diagnosis model based on multiple fault diagnosis results. The voting factor of each fault type is obtained by testing multiple fault diagnosis models using a test set. The test set includes test samples and the actual fault types corresponding to each test sample. For any fault diagnosis model, the voting factor of each fault type is calculated based on the prediction precision of the fault diagnosis model for each fault type, the prediction recall of the fault diagnosis model for each fault type, and the average confidence of multiple fault diagnosis models for each fault type. The decision weight determination module is used to determine the decision weight of each fault diagnosis model; the decision weight is obtained by testing multiple fault diagnosis models using the test set; The decision threshold determination module is used to determine the decision threshold based on the voting factors of various fault types obtained by testing each fault diagnosis model using the test set; The diagnostic result fusion module is used to perform multi-level decision fusion on the fault diagnosis results output by multiple fault diagnosis models based on the voting factor of the fault type of the target battery output by each fault diagnosis model, the decision weight of each fault diagnosis model and the decision threshold, and select the final fault diagnosis result of the target battery. Determine the decision weights for each fault diagnosis model, specifically including: For any fault diagnosis model, based on the actual fault type and predicted fault type of each test sample output by the fault diagnosis model, determine the number of correctly classified test samples and the number of incorrectly classified test samples. For any fault diagnosis model, the ratio of the number of correctly classified test samples to the total number of test samples is determined as the decision weight of the fault diagnosis model. Among them, the Each fault diagnosis model outputs the fault type. Credibility The calculation formula is: ； Indicates the fault type The number of test samples, Indicates that the test set is input to the first... Each fault diagnosis model is classified into fault types. The number of test samples; No. Each fault diagnosis model outputs the fault type. Average credibility The calculation formula is: ； in, m This indicates the number of fault diagnosis models.
7. An electronic device, characterized in that, The device includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to cause the electronic device to perform the electric vehicle power battery fault diagnosis method according to any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the electric vehicle power battery fault diagnosis method as described in any one of claims 1 to 5.