A method for predicting battery insulation failure in electric vehicles

Abstract

The present invention provides a method for predicting battery insulation failure of electric vehicles, comprising: a machine learning platform merges and segments vehicle historical data obtained from a big data platform to obtain at least two types of battery insulation failure impact data; the machine learning platform uses the battery insulation failure impact data to train a preset linear classifier to obtain a battery insulation failure prediction model; a business operation platform applies the battery insulation failure prediction model sent by the machine learning platform to perform offline battery insulation failure risk prediction based on the current battery insulation failure impact data of the vehicle side; alternatively, the vehicle side applies the battery insulation failure prediction model to perform online battery insulation failure risk prediction based on its current battery insulation failure impact data; the present invention can realize real-time online battery insulation failure risk prediction on the vehicle side and offline battery insulation failure risk prediction on the business operation platform, thereby improving the reliability of battery insulation failure risk prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of battery failure prediction for electric vehicles, and in particular to a method for predicting battery insulation failure for electric vehicles. Background Art

[0002] With the rapid development of electric vehicles in recent years, their penetration rate has increased significantly. Electric vehicle safety accidents are also common, with the increasing number of accidents caused by power batteries, leading to increasing attention to the safety of power batteries. Currently, existing solutions for battery insulation failure early warning generally rely on analyzing battery data such as battery temperature, battery voltage, and battery insulation internal resistance. These solutions fail to consider the impact of driving habits, vehicle operating conditions, and the external environment on power battery insulation failure, resulting in poor reliability in power battery insulation failure prediction results. Summary of the Invention

[0003] In response to the above problems, the purpose of the present invention is to provide a battery insulation failure prediction method for electric vehicles, which can realize real-time online battery insulation failure risk prediction on the vehicle side and offline battery insulation failure risk prediction on the business operation platform, thereby improving the reliability of battery insulation failure risk prediction.

[0004] An embodiment of the present invention provides a method for predicting battery insulation failure of an electric vehicle, comprising:

[0005] The machine learning platform merges and segments the historical vehicle data obtained from the big data platform to obtain at least two types of battery insulation failure impact data;

[0006] The machine learning platform uses the battery insulation failure impact data to train a preset linear classifier to obtain a battery insulation failure prediction model;

[0007] The business operation platform applies the battery insulation failure prediction model sent by the machine learning platform to perform offline battery insulation failure risk prediction based on the current battery insulation failure impact data of the vehicle side; alternatively, the vehicle side applies the battery insulation failure prediction model to perform online battery insulation failure risk prediction based on its current battery insulation failure impact data.

[0008] As an improvement to the above solution, the machine learning platform uses the battery insulation failure impact data to train a preset linear classifier to obtain a battery insulation failure prediction model, including:

[0009] The machine learning platform slices the battery insulation failure impact data to obtain a plurality of data slices; wherein each data slice has an equal length;

[0010] The machine learning platform trains a preset linear classifier based on the data slices to obtain a battery insulation failure prediction model.

[0011] As an improvement to the above solution, the machine learning platform merges and segments the vehicle historical data obtained from the big data platform to obtain at least two types of battery insulation failure impact data, including:

[0012] The machine learning platform merges the vehicle historical data according to the vehicle to which it belongs and the timestamp;

[0013] The machine learning platform divides the merged vehicle historical data according to vehicle status to obtain at least two types of battery insulation failure impact data.

[0014] As an improvement to the above solution, the vehicle status includes a high voltage state and a charging state of the battery;

[0015] Then, the machine learning platform segments the combined vehicle historical data according to vehicle status to obtain at least two types of battery insulation failure impact data, including:

[0016] The machine learning platform divides the merged vehicle historical data into battery insulation failure impact data corresponding to the high voltage state on the battery and battery insulation failure impact data corresponding to the charging state.

[0017] As an improvement to the above solution, the method further includes:

[0018] Before slicing, the machine learning platform performs dimensionality reduction processing on the battery insulation failure impact data;

[0019] After slicing, the machine learning platform performs standardization and normalization on the data slices.

[0020] As an improvement to the above solution, the method further includes:

[0021] The machine learning platform performs data cleaning and / or data missing filling processing on the battery insulation failure impact data after dimensionality reduction.

[0022] As an improvement to the above solution, the machine learning platform performs dimensionality reduction processing on the battery insulation failure impact data, including:

[0023] The machine learning platform performs a Pearson correlation coefficient analysis on the battery insulation failure impact data, and selects battery insulation failure impact data with a Pearson coefficient less than a first set value as the battery insulation failure impact data after dimensionality reduction;

[0024] Alternatively, the machine learning platform performs PCA dimensionality reduction on the battery insulation failure impact data to obtain the battery insulation failure impact data after dimensionality reduction.

[0025] As an improvement to the above solution, the data missing filling process specifically includes:

[0026] When one of the data in the battery insulation failure impact data is event-type data, the data corresponding to the previous timestamp is used to fill in the missing data;

[0027] When one of the battery insulation failure impact data is continuous periodic data, the data is filled with missing data using an average value filling method;

[0028] When one of the battery insulation failure impact data is discrete periodic data, the data corresponding to the previous timestamp is used to perform missing filling on the data.

[0029] As an improvement to the above solution, the battery insulation failure impact data includes driving behavior data, vehicle data, vehicle operating condition data, battery data and environmental data.

[0030] As an improvement to the above solution, the data cleaning specifically includes:

[0031] The 3σ principle of normal distribution is used to perform data cleaning on the battery insulation failure influencing data after dimensionality reduction, and invalid data and duplicate data in the battery insulation failure influencing data are eliminated.

[0032] As an improvement to the above solution, the method includes:

[0033] The business operation platform performs monitoring and early warning based on the battery insulation failure risk predicted by itself or the battery insulation failure risk predicted by the vehicle side;

[0034] Alternatively, the user end or the vehicle system issues an early warning based on the battery insulation failure risk predicted by the vehicle end.

[0035] Compared with the prior art, the beneficial effects of the embodiments of the present invention are: by acquiring vehicle historical data pre-stored in a big data platform; merging and segmenting the vehicle historical data to obtain at least two types of battery insulation failure impact data; preprocessing the battery insulation failure impact data, and using the battery insulation failure impact data to train a preset linear classifier to obtain a battery insulation failure prediction model; the vehicle side applies the battery insulation failure prediction model to perform online battery insulation failure risk prediction based on its current battery insulation failure impact data; at the same time, the business operation platform applies the battery insulation failure prediction model sent by the machine learning platform to perform offline battery insulation failure risk prediction based on the current battery insulation failure impact data of the vehicle side; the present invention uses a large amount of vehicle historical data of people, vehicles and environments pre-stored on a big data platform to perform machine learning, obtain a battery insulation failure prediction model and send it to the vehicle side and the business operation platform, realizing real-time online battery insulation failure risk prediction of the vehicle side and offline battery insulation failure risk prediction of the business operation platform, thereby improving the reliability of battery insulation failure risk prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the implementation. Obviously, the drawings described below are only some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0037] Figure 1 This is a flow chart of a method for predicting battery insulation failure of an electric vehicle provided by an embodiment of the present invention;

[0038] Figure 2 This is a schematic block diagram of battery insulation failure prediction provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0040] See also Figure 1 The present invention provides a method for predicting battery insulation failure of an electric vehicle, comprising:

[0041] S1: The machine learning platform merges and segments the historical vehicle data obtained from the big data platform to obtain at least two types of battery insulation failure impact data;

[0042] Furthermore, the battery insulation failure impact data includes driving behavior data, vehicle data, vehicle operating condition data, battery data and environmental data.

[0043] For example, each control domain of the vehicle, such as the VDCM domain, BDCM domain, and IDVM domain, uploads its current domain data, such as VDCM domain data, BDCM domain data, and IDVM domain data, to the big data platform in real time through the on-board gateway T-BOX. That is, the vehicle historical data stored in the big data platform includes the data uploaded by each control domain on the vehicle side. The big data platform analyzes and preliminarily screens the real-time domain data uploaded by the vehicle controller, divides them into five categories of data, and stores them in different data sources, including whole vehicle data, vehicle operating condition data, battery data, driving behavior data, and environmental data. For example, the whole vehicle data includes whole vehicle parameters, power, torque, vehicle weight, shock absorber current, motor status, motor current / motor voltage, etc.; the vehicle operating condition data includes tire pressure, tire temperature, interior temperature, air-conditioning power, six doors and two lids, lights, motor speed, speed, etc.; the battery data includes current, voltage, battery temperature, SOC, fast charging, slow charging, battery characteristics during driving, battery characteristics during charging, etc.; the driving behavior data includes vehicle speed, wheel speed, air-conditioning control, brake pedal force, seat belt control, light control, parking, gear, etc.; the environmental data includes weather information such as rainfall, weather, temperature, humidity, road condition information such as mountainous areas, cities, highways, speed limits, etc., and can also include charging pile information during charging. Initial screening of the real-time data can be performed based on its impact or relevance to vehicle range, eliminating data irrelevant to range estimation. This can be determined based on expert experience. The big data platform stores over 200 different types of signal data for both the same and different vehicles, providing a massive data foundation for subsequent range predictions. Each historical vehicle data entry carries a timestamp and vehicle status information.

[0044] It should be noted that the processing of uploaded vehicle historical data by the big data platform belongs to the existing technology and will not be described in detail here. Through the big data platform, different vehicle historical data can be saved to different data sources, for example, according to static data, dynamic data, header structured data, semi-structured data, and unstructured data.

[0045] Since the endurance is time series data, in the embodiment of the present invention, the machine learning platform extracts the vehicle historical data from the big data platform in the order of timestamps and constructs the model.

[0046] Further, the vehicle historical data is screened;

[0047] Exemplarily, after determining the data magnitude, the vehicle historical data of N vehicles within a set time period is screened out, for example, 90 days of vehicle historical data of 100 vehicles are extracted from the big data platform in chronological order of timestamps as the initial data set.

[0048] Furthermore, S1: the machine learning platform merges and segments the vehicle historical data obtained from the big data platform to obtain at least two types of battery insulation failure impact data, including:

[0049] The machine learning platform merges the vehicle historical data according to the vehicle to which it belongs and the timestamp;

[0050] The following table provides an example of the combined vehicle history data.

[0051] Frame number Timestamp Data 1 Data 2 Data 3 Data 4 ... Data M Remark Vehicle 1 1633609494000 3 0 444.5 72.1 ... 2 Hibernation Vehicle 1 1633609494000 3 0 444.5 72.1 ... 2 Fast charging Vehicle 2 1633609495000 3 0 369 72.1 ... 0 slow charge Vehicle 2 1633609496000 2 0 78.75 72.1 ... 0 Upper high voltage ... ... ... ... ... ... ... ... ... Vehicle N 1633609497000 2 0 14 72.1 ... 0 Upper high voltage

[0052] The machine learning platform divides the merged vehicle historical data according to vehicle status to obtain at least two types of battery insulation failure impact data.

[0053] The vehicle status includes the high-voltage state on the battery (i.e., the driving state), the charging state (including fast charging and slow charging), the dormant state, etc. Since the factors affecting the endurance will be different when the vehicle is in different states, therefore, when segmenting the battery insulation failure impact data, it is necessary to segment the appropriate data from the vehicle historical data according to the vehicle states corresponding to different timestamps. For example, during the charging process, data such as speed and acceleration cannot be used as battery insulation failure impact data. During driving, charging piles and discharge guns cannot be used as battery insulation failure impact data. Through data merging and segmentation processing, the vehicle historical data can be divided into different types of battery insulation failure impact data, so that the prediction of the cruising range is more in line with the current state of the vehicle.

[0054] S2: The machine learning platform uses the battery insulation failure impact data to train a preset linear classifier to obtain a battery insulation failure prediction model;

[0055] S3: The business operation platform applies the battery insulation failure prediction model sent by the machine learning platform to perform offline battery insulation failure risk prediction based on the current battery insulation failure impact data on the vehicle side; alternatively, the vehicle side applies the battery insulation failure prediction model to perform online battery insulation failure risk prediction based on its current battery insulation failure impact data.

[0056] The embodiment of the present invention uses a big data platform to pre-store a large amount of vehicle historical data of people, vehicles and environments for machine learning, obtains a battery insulation failure prediction model and sends it to the vehicle side, so that the vehicle side performs online edge computing through the battery insulation failure prediction model to obtain the battery insulation failure risk corresponding to its current battery insulation failure impact data. At the same time, the business operation platform applies the battery insulation failure prediction model sent by the machine learning platform to perform offline battery insulation failure risk prediction based on the current battery insulation failure impact data of the vehicle side, thereby realizing real-time online battery insulation failure risk prediction on the vehicle side and offline battery insulation failure risk prediction on the business operation platform, thereby improving the reliability of battery insulation failure risk prediction.

[0057] In an optional embodiment, the vehicle state includes a high voltage state and a charging state of the battery;

[0058] Then, the merged vehicle historical data is segmented according to the vehicle status to obtain at least two types of battery insulation failure impact data, including:

[0059] The combined vehicle history data is divided into battery insulation failure impact data corresponding to the high voltage state of the battery and battery insulation failure impact data corresponding to the charging state.

[0060] In an optional embodiment, the machine learning platform uses the battery insulation failure impact data to train a preset linear classifier to obtain a battery insulation failure prediction model, including:

[0061] The machine learning platform slices the battery insulation failure impact data to obtain a plurality of data slices; wherein each data slice has an equal length;

[0062] The machine learning platform trains a preset linear classifier based on the data slices to obtain a battery insulation failure prediction model.

[0063] In an embodiment of the present invention, data is sliced ​​using a set time interval. For example, assuming there are 1 million data items, with one data item per second, slicing with a 30s interval yields more than 30,000 data slices. Slicing with a 300s interval yields more than 30,000 data slices, and more than 3,000 data slices. Each data slice is used as a sample and input into a preset linear classifier, such as an SVM classifier, for training.

[0064] Furthermore, to improve data validity, interval filtering is also required. Specifically, by traversing all intervals, if the amount of data in the next interval is less than a set number, the interval is eliminated. The value of the set number is equal to the value of the set duration corresponding to the interval. That is, the length of the data slices after interval filtering is equal.

[0065] In an optional embodiment, the method further includes:

[0066] Before slicing, the machine learning platform performs dimensionality reduction processing on the battery insulation failure impact data;

[0067] For example, the battery insulation failure impact data may be subjected to dimensionality reduction processing using methods such as mean, median, standard deviation, data visualization, histogram, Pearson correlation coefficient, and PAC dimensionality reduction to eliminate data that is not strongly correlated with battery life.

[0068] Furthermore, the machine learning platform performs dimensionality reduction processing on the battery insulation failure impact data, including:

[0069] The machine learning platform performs a Pearson correlation coefficient analysis on the battery insulation failure impact data, and selects battery insulation failure impact data with a Pearson coefficient less than a first set value as the battery insulation failure impact data after dimensionality reduction;

[0070] Alternatively, the machine learning platform performs PCA dimensionality reduction on the battery insulation failure impact data to obtain the battery insulation failure impact data after dimensionality reduction.

[0071] It should be noted that the Pearson correlation coefficient is defined as the quotient of the covariance and standard deviation between two data, which expresses the coefficient of linear correlation; among them, there is a linear relationship between the data, and they are all continuous data; the data population is normally distributed, or a unimodal distribution close to normal; the observations of the data are paired, and each pair of observations is independent of each other.

[0072] By calculating the probability value that there is no correlation between the two data, that is, the Pearson coefficient P value, and then calculating the coefficient, that is, the R value;

[0073] If the Pearson coefficient P value is very small, the R value is valid, indicating that the two data are correlated, and the two data are retained. For example, the first set value is set to 0.05 and the second set value is set to 0.5. This can filter out data with a Pearson coefficient P value > 0.5 and a P value < 0.05, achieving data dimensionality reduction, obtaining data with a P value < 0.05, and eliminating weakly correlated or unrelated data.

[0074] In other embodiments, each collected data point can be analyzed using the mean, median, and mode, respectively, for each unit. For example, the average temperature can be used for vehicle interior temperature and tire pressure, the mode can be used for resistance, and the median can be used for speed. By analyzing each data point and determining the difference between the changes in these values ​​and the actual values, some data with poor quality can be excluded, achieving a partial dimensionality reduction effect. For example, in the analysis of remote control T-Box power data, it was found that none of the mean, mode, or median values ​​matched actual operation, indicating data quality issues and thus failing to meet the requirements as a feature, and thus this data was eliminated.

[0075] Among them, the PAC dimension reduction principle is as follows:

[0076] Suppose there are m pieces of n-dimensional battery insulation failure impact data.

[0077] The original battery insulation failure impact data is organized into an n-row and m-column matrix X;

[0078] Zero-mean each row of X, that is, subtract the mean of this row, and then calculate the covariance matrix to calculate the eigenvalues ​​and corresponding eigenvectors of the covariance matrix;

[0079] The eigenvectors are arranged into a matrix by row from top to bottom according to the corresponding eigenvalue size, and the first k rows are taken to form the matrix P, which is the battery insulation failure impact data after dimensionality reduction to k dimensions.

[0080] After slicing, the machine learning platform performs standardization and normalization on the data slices.

[0081] In the embodiment of the present invention, the problem of dimensionless data attributes can be solved by standardizing and normalizing the battery insulation failure impact data. Specifically, through the formula Standardization is performed by formula Normalization is performed. Where μ represents the average value of the battery insulation failure impact data, σ represents the standard deviation of the battery insulation failure impact data, X represents a battery insulation failure impact data, X min Indicates the minimum value of the battery insulation failure impact data, X max Indicates the maximum value of the battery insulation failure impact data.

[0082] In an optional embodiment, the method further includes:

[0083] The machine learning platform performs data cleaning and / or data missing filling processing on the battery insulation failure impact data after dimensionality reduction.

[0084] Exemplarily, data cleaning can be performed using the 3σ principle of normal distribution, value range, NA cleaning, deduplication rules, etc. In an embodiment of the present invention, the 3σ principle of normal distribution is preferably used for data cleaning.

[0085] Furthermore, the data cleaning specifically includes:

[0086] The 3σ principle of normal distribution is used to perform data cleaning on the battery insulation failure influencing data after dimensionality reduction, and invalid data and duplicate data in the battery insulation failure influencing data are eliminated.

[0087] For example, based on the 3σ principle of normal distribution, values ​​that deviate from the mean μ by more than 3 times the standard deviation σ can be screened out. If the data obeys the normal distribution and the distance from the mean is 3σ, the probability of the occurrence of a value is P(|x-μ|>3σ)<=0.003, which is a low-probability event, and this part of the data is eliminated.

[0088] The data missing filling process is specifically processed;

[0089] When one of the data in the battery insulation failure impact data is event-type data, the data corresponding to the previous timestamp is used to fill in the missing data;

[0090] When one of the battery insulation failure impact data is continuous periodic data, the data is filled with missing data using an average value filling method;

[0091] When one of the battery insulation failure impact data is discrete periodic data, the data corresponding to the previous timestamp is used to perform missing filling on the data.

[0092] Since data is collected in timestamp order at a specified frequency, data loss and incorrect data collection may occur during the collection process due to influences such as the network, sensors, and performance, resulting in invalid data. Therefore, corresponding event-type missing data can be filled with the data corresponding to the previous timestamp, continuous periodic missing data can be filled with the mean, and discrete periodic data can be filled with the data of the previous timestamp to ensure that the data sequence remains valid.

[0093] In an optional embodiment, the linear classifier is a SVM classifier;

[0094] Then, the machine learning platform trains a preset linear classifier according to the data slices to obtain a battery insulation failure prediction model, including:

[0095] Dividing the data slices into a training sample set, a validation sample set, and a test sample set;

[0096] The data slices in the training sample set are sequentially used as input feature data of the SVM classifier, and the battery insulation failure risk corresponding to the data slices is used as the output feature of the SVM classifier to construct the battery insulation failure prediction model; wherein the battery insulation failure risk includes: failure and normal;

[0097] Using the verification sample set to perform model verification on the battery insulation failure prediction model;

[0098] The test sample set is used to perform model testing on the battery insulation failure prediction model.

[0099] In the embodiment of the present invention, the SVM classifier uses a ploy kernel (ie, a polynomial kernel) to construct a linear function.

[0100] In an embodiment of the present invention, the data is divided into a training set, a validation set, and a test set according to a set ratio, for example, 60%, 20%, and 20%, and model training, validation, and testing are performed in sequence to ensure that the prediction accuracy of the constructed battery insulation failure prediction model meets expectations.

[0101] After building the battery insulation failure prediction model, it's transmitted to the vehicle-side domain controller. The vehicle-side domain controller uses the model and the currently collected battery insulation failure impact data to predict battery insulation failure. Since the construction and subsequent iterative updates of the battery insulation failure impact data don't need to be performed on the vehicle, the vehicle only needs to perform edge computing, effectively reducing the amount of computation required and improving the efficiency of vehicle-side range prediction.

[0102] In an optional embodiment, the method includes:

[0103] The business operation platform performs monitoring and early warning based on the battery insulation failure risk predicted by itself or the battery insulation failure risk predicted by the vehicle side;

[0104] Alternatively, the user end or the vehicle system issues an early warning based on the battery insulation failure risk predicted by the vehicle end.

[0105] like Figure 2As shown, after constructing a battery insulation failure prediction model, the machine learning platform sends it to the business operations platform, enabling the platform to use it to predict offline battery insulation failure risks, thereby enabling batch offline monitoring and early warning on the vehicle side. Simultaneously, the business operations platform sends the model to the vehicle side via the TSP cloud platform and T-BOX, enabling the vehicle side to use it to predict online battery insulation failure risks. After the vehicle side predicts the battery insulation failure risk, it uploads it to its vehicle-mounted system via T-BOX for early warning. It also uploads it to the user-side app via T-BOX and the TSP cloud platform for early warning. Furthermore, the vehicle side also sends the battery insulation failure risk to the business operations platform via T-BOX and the TSP cloud platform for monitoring and early warning. Simultaneously, the business operations platform monitors and warns of its own predicted battery insulation failure risk, thereby enabling simultaneous early warning of battery insulation failure risks on the vehicle side and the business operations platform.

[0106] Compared with the prior art, the beneficial effects of the embodiments of the present invention are: by acquiring vehicle historical data pre-stored in a big data platform; merging and segmenting the vehicle historical data to obtain at least two types of battery insulation failure impact data; using the battery insulation failure impact data to train a preset linear classifier to obtain a battery insulation failure prediction model; the vehicle side applies the battery insulation failure prediction model to perform online battery insulation failure risk prediction based on its current battery insulation failure impact data; at the same time, the business operation platform applies the battery insulation failure prediction model sent by the machine learning platform to perform offline battery insulation failure risk prediction based on the current battery insulation failure impact data of the vehicle side; the present invention uses a big data platform to pre-store a large amount of vehicle historical data of people, vehicles and environments for machine learning, obtains a battery insulation failure prediction model and sends it to the vehicle side and the business operation platform, realizing real-time online battery insulation failure risk prediction of the vehicle side and offline battery insulation failure risk prediction of the business operation platform, thereby improving the reliability of battery insulation failure risk prediction.

[0107] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for predicting battery insulation failure of an electric vehicle, characterized in that: The machine learning platform merges and segments historical vehicle data obtained from the big data platform to obtain at least two types of battery insulation failure impact data; the battery insulation failure impact data includes driving behavior data, vehicle data, vehicle operating condition data, battery data, and environmental data; The machine learning platform uses the battery insulation failure impact data to train a preset linear classifier to obtain a battery insulation failure prediction model, including: the machine learning platform slices the battery insulation failure impact data to obtain a plurality of data slices; wherein each data slice has an equal length; the machine learning platform trains a preset linear classifier based on the data slices to obtain a battery insulation failure prediction model; Before slicing, the machine learning platform performs dimensionality reduction processing on the battery insulation failure influencing data; after slicing, the machine learning platform performs standardization and normalization processing on the data slices; the machine learning platform performs dimensionality reduction processing on the battery insulation failure influencing data, including: the machine learning platform performs Pearson correlation coefficient analysis on the battery insulation failure influencing data, and screens out battery insulation failure influencing data with a Pearson coefficient less than a first set value as the battery insulation failure influencing data after dimensionality reduction; or, the machine learning platform performs PCA dimensionality reduction on the battery insulation failure influencing data to obtain the battery insulation failure influencing data after dimensionality reduction; The business operation platform applies the battery insulation failure prediction model sent by the machine learning platform to perform offline battery insulation failure risk prediction based on the current battery insulation failure impact data of the vehicle side; alternatively, the vehicle side applies the battery insulation failure prediction model to perform online battery insulation failure risk prediction based on its current battery insulation failure impact data.

2. The method for predicting battery insulation failure of an electric vehicle according to claim 1, wherein: The machine learning platform merges and segments the historical vehicle data obtained from the big data platform to obtain at least two types of battery insulation failure impact data, including: The machine learning platform merges the vehicle historical data according to the vehicle to which it belongs and the timestamp; The machine learning platform divides the merged vehicle historical data according to vehicle status to obtain at least two types of battery insulation failure impact data.

3. The method for predicting battery insulation failure of an electric vehicle according to claim 2, wherein: The vehicle status includes the high voltage status and charging status of the battery; Then, the machine learning platform segments the combined vehicle historical data according to vehicle status to obtain at least two types of battery insulation failure impact data, including: The machine learning platform divides the merged vehicle historical data into battery insulation failure impact data corresponding to the high voltage state on the battery and battery insulation failure impact data corresponding to the charging state.

4. The method for predicting battery insulation failure of an electric vehicle according to claim 1, wherein: The method further comprises: The machine learning platform performs data cleaning and / or data missing filling processing on the battery insulation failure impact data after dimensionality reduction.

5. The method for predicting battery insulation failure of an electric vehicle according to claim 4, wherein: The data missing filling process specifically includes: When one of the data in the battery insulation failure impact data is event-type data, the data corresponding to the previous timestamp is used to fill in the missing data; When one of the battery insulation failure impact data is continuous periodic data, the data is filled with missing data using an average value filling method; When one of the battery insulation failure impact data is discrete periodic data, the data corresponding to the previous timestamp is used to perform missing filling on the data.

6. The method for predicting battery insulation failure of an electric vehicle according to claim 4, wherein: The data cleaning specifically includes: The 3σ principle of normal distribution is used to perform data cleaning on the battery insulation failure influencing data after dimensionality reduction, and invalid data and duplicate data in the battery insulation failure influencing data are eliminated.

7. The method for predicting battery insulation failure of an electric vehicle according to claim 1, wherein: The method includes: the business operation platform monitors and issues early warnings based on the battery insulation failure risk predicted by itself or the battery insulation failure risk predicted by the vehicle end; or, the user end or the vehicle system issues early warnings based on the battery insulation failure risk predicted by the vehicle end.

Images