Battery charging abnormal state monitoring method and monitoring system based on machine learning

Abstract

The application discloses a battery charging abnormal state monitoring method and system based on machine learning, and belongs to the technical field of heavy truck charging stations, and is characterized in that the method comprises the following steps: S1: collecting historical charging data of a battery; S2: preprocessing the historical charging data; S3: constructing a feature set; S4: performing data normalization processing on the feature set, and scaling all numerical values to the range of [0, 1]; S5: utilizing a PCA algorithm to reduce dimensions, and reducing the feature set from 2N+6 to K dimensions; S6: utilizing a KNN algorithm to fit the feature set after dimension reduction, and obtaining a trained clustering model and a clustering center point; and S7: inputting the feature set after dimension reduction into the clustering model, and outputting the Euclidean distance of each feature point to the clustering center point, wherein the greater the Euclidean distance value is, the higher the abnormality degree of the feature point is. Through analysis of charging historical big data, the application can realize intelligent analysis of the abnormal state of the battery.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of heavy truck charging stations, and particularly relates to a battery charging abnormal state monitoring method and monitoring system based on machine learning. BACKGROUND

[0002] In recent years, with the rapid development of new energy technology, the number of heavy truck battery swap stations is increasing. Unlike traditional household car charging stations, heavy truck battery swap stations are used to charge the battery packs of heavy trucks after disassembly. In the charging process, if a fire accident occurs, it may cause the entire heavy truck battery swap station to be destroyed. Therefore, the charging safety of heavy truck battery swap stations needs to be paid great attention to. At present, the existing heavy truck battery swap station carriers monitor the abnormality in two main ways during the battery charging process:

[0003] 1. The station communicates with the BMS to obtain the battery current, temperature, voltage and other values in real time, and compares them with the threshold values set in advance. If the threshold values are exceeded, an alarm is given, such as high temperature, excessive current, excessive voltage, etc.

[0004] 2. The station communicates with the BMS to obtain the alarm data generated by the BMS system in real time.

[0005] Through practice, it is found that the above two methods can monitor the battery abnormality to some extent, but still have limitations. First, this type of method can only monitor the abnormal types determined in advance, and cannot monitor some unknown potential abnormalities (such as voltage and current correlation) because they are not set in advance. Second, the monitoring standard lacks individualization and does not distinguish between different batteries, such as using the same abnormal monitoring threshold for batteries of different ages. SUMMARY

[0006] In view of the defects of the prior art, the application provides a battery charging abnormal state monitoring method and monitoring system based on machine learning, which realizes intelligent analysis of the battery abnormal state through analysis of the charging history big data.

[0007] The specific technical scheme adopted by the application is as follows:

[0008] The first invention of the patent is to provide a battery charging abnormal state monitoring method based on machine learning, which comprises the following steps:

[0009] S1: Collecting historical charging data of the battery: Specifically, according to the data sampling frequency, M rows of data fields are extracted, and each row of data field includes battery box code, temperature, voltage, current, soc, start charging time and data timestamp; M is a natural number greater than 1;

[0010] S2, preprocessing the historical charging data, specifically:

[0011] S2.1, start charging time, data packet, each group of historical charging data represents all data of a piece of battery once charging process by battery box coding;

[0012] S2.2, sort each group of historical charging data from low to high according to timestamp;

[0013] S2.3, screen out data groups with starting soc less than 40 and ending soc greater than 98;

[0014] S2.4, for each group of historical charging data, screen out data rows in the period from soc 40 to 98;

[0015] S3, construct feature set; specifically:

[0016] S3.1, for each group of historical charging data, construct the following features: current value standard deviation, current curve single-point fluctuation number, current curve step fluctuation number, interpolated current sequence I' N , interpolated voltage sequence U' N , maximum temperature, minimum temperature and average temperature;

[0017] S3.2, by constructing features, each group of charging historical data is converted into a 1x(2N+6) feature row; get m x (2N+6) feature set, m is the number of data groups after data preprocessing;

[0018] S4, normalize the feature set, scale all numerical values to the range [0, 1];

[0019] S5, use PCA algorithm for dimension reduction, reduce the feature set from 2N+6 to K dimensions;

[0020] S6, use KNN algorithm to fit the dimension-reduced feature set to get the trained clustering model and clustering center point;

[0021] S7, input the dimension-reduced feature set into the trained clustering model, output the Euclidean distance from each feature point to the clustering center point, the greater the Euclidean distance value, the higher the abnormality of the feature point. Wherein: each feature point represents a row of feature data of s3.2.

[0022] Further: the data sampling frequency is 6 times per minute.

[0023] Further: in S3.1,

[0024] The formula for obtaining the current value standard deviation is:

[0025] Take the current sequence I n in the group data, the number of group data is n, then the current average value I μFor:

[0026]

[0027] current standard deviation I σ For:

[0028]

[0029] the current value of the battery at a certain moment is I i , the current value at the previous moment is I i-1 , the current value at the next moment is I i+1 , and the set threshold value is T.

[0030] The acquisition process of the number of single-point fluctuations of the current curve is: if (I i-1 -I i >T)&(I i+1 -I i >T), then I i is determined as a single-point fluctuation of the current curve, and the count is 1. For each set of historical charging data, the total number is calculated, which is m1.

[0031] The acquisition process of the number of step fluctuations of the current curve is: if (|I i-1 -I i |≤T)&(|I i+1 -I i |≤T), then I i is determined as a continuous change point. After removing the single-point fluctuation points and the continuous change points, for the remaining points, if (I

[0032] (I i -I i+1 >T)&(|I i+1 -I i+2 |≤T)&(I i+3 -I i+2 >T), then I i is determined as a step fluctuation of the current curve, and the count is 1. For each set of historical charging data, the total number is calculated, which is n2.

[0033] interpolated current sequence I' N interpolated voltage sequence U' N Take the current sequence I n , the voltage sequence U n , and the data timestamp sequence Time n of each set of historical charging data. First, calculate the difference between each time point and the starting time point to obtain the second count time sequence:

[0034] t i = Time i -Time1(1≤i≤n,i∈Z).

[0035] Take the interpolation number N, and find the interpolation time sequence:

[0036]

[0037] Then the interpolation current sequence is:

[0038]

[0039] Then the interpolation voltage sequence is:

[0040]

[0041] Further: in S3.2: perform S3.1 on each set of historical charging data, finally obtain an m x (2N+6) feature set, m is the number of data sets after preprocessing of the original data, and 2N+6 is the number of features constructed by each set of historical charging data; for each column f i The following normalization processing is performed:

[0042] f max = max{f i}(1≤i≤m,i∈Z);

[0043] f min = min{f i}(1≤i≤m,i∈Z);

[0044] f i = (f i -f min ) / (f max -f min )(1≤i≤m,i∈Z);

[0045] Obtain the final feature set F.

[0046] The second invention of the patent is to provide a battery charging abnormal state monitoring system based on machine learning, comprising:

[0047] The basic data collection module collects the historical charging data of the battery: specifically, according to the data sampling frequency, M rows of data fields are extracted, each row of data field includes battery box code, temperature, voltage, current, soc, start charging time, data timestamp; M is a natural number greater than 1;

[0048] The preprocessing module pre-processes the historical charging data, specifically:

[0049] S2.1, group data by battery box code and start charging time, and each set of historical charging data represents all data of a charging process of a battery;

[0050] S2.2, sort each set of historical charging data from low to high by timestamp;

[0051] S2.3, filter out the data set whose starting soc is less than 40 and ending soc is greater than 98;

[0052] S2.4, for each set of historical charging data, filter out the data row in the time period from soc 40 to 98;

[0053] Construction module: construct feature set; specifically:

[0054] S3.1, for each set of historical charging data, construct the following features: current value standard deviation, number of single-point fluctuations in current curve, number of step fluctuations in current curve, interpolated current sequence I' N , interpolated voltage sequence U' N , maximum temperature value, minimum temperature value and average temperature value;

[0055] S3.2, through the construction of features, each set of charging history data is converted into a 1x(2N+6) feature row; obtain a feature set of m x (2N+6), m is the number of data sets after data preprocessing;

[0056] Normalization processing module: perform data normalization processing on the feature set, and scale all numerical values to the range [0, 1];

[0057] Dimensionality reduction module: use PCA algorithm for dimensionality reduction, reduce the feature set from 2N+6 to K dimensions;

[0058] Model construction module: use KNN algorithm to fit the dimensionality-reduced feature set to obtain the trained clustering model and clustering center point;

[0059] Abnormal analysis module: input the dimensionality-reduced feature set into the clustering model to obtain.

[0060] Further: the data sampling frequency is 6 times per minute.

[0061] Further: in S3.1,

[0062] The formula for obtaining the current value standard deviation is:

[0063] Take the current sequence I n in the data set, and the number of data sets is n, then the current average value I μ is:

[0064]

[0065] The current standard deviation I σ is:

[0066]

[0067] The current value of the battery at a certain moment during charging is I i The current value at the previous moment was I i-1 The current value at the next moment is I i+1 Set the threshold to T;

[0068] The process for obtaining the number of single-point fluctuations in the current curve is as follows: If (I i-1 -I i >T)&(I i+1 -I i >T), then I i If the current curve is determined to be a single-point fluctuation, the count is 1. For each set of historical charging data, the total number is calculated and counted as n1.

[0069] The process for obtaining the number of step fluctuations in the current curve is as follows: If (|I i-1 -I i |≤T)&(|I i+1 -I i If |≤T), then I i Determine points of continuous change; after removing single-point fluctuation points, remove points of continuous change. For the remaining points, if...

[0070] (I i -I i+1 >T)&(|I i+1 -I i+2 |≤T)&(I i+3 -I i+2 >T), then I i If the current curve is identified as a stepped fluctuation, it is counted as 1. For each set of historical charging data, the total number is calculated and counted as n2.

[0071] Interpolated current sequence I′ N Interpolated voltage sequence U′ N Take the current sequence I of each group of historical charging data n Voltage sequence U n Data timestamp sequence Time n First, calculate the difference between each time point and the starting time point to obtain the second count time series:

[0072] t i =Time i -Time1(1≤i≤n,i∈Z);

[0073] Take the number of interpolations N, and calculate the interpolation time series:

[0074]

[0075] The interpolated current sequence is then:

[0076]

[0077] The interpolation voltage sequence is:

[0078]

[0079] Further, in S3.2: perform S3.1 on each set of historical charging data, finally obtain a feature set of m x (2N+6), m is the number of data sets after preprocessing of the original data, (2N+6) is the number of features constructed by each set of historical charging data; for each column f i The following normalization processing is performed:

[0080] f max = max{f i}(1≤i≤m,i∈Z);

[0081] f min = min{f i}(1≤i≤m,i∈Z);

[0082] f i = (f i -f min ) / (f max -f min )(1≤i≤m,i∈Z);

[0083] Obtain the final feature set F.

[0084] The third inventive purpose of the patent is to provide a computer program for implementing the above-mentioned machine learning-based battery charging abnormal state monitoring method.

[0085] The fourth inventive purpose of the patent is to provide an information data processing terminal for implementing the above-mentioned machine learning-based battery charging abnormal state monitoring method.

[0086] The fifth inventive purpose of the patent is to provide a computer readable storage medium comprising instructions which, when executed on a computer, cause the computer to perform the above-mentioned machine learning-based battery charging abnormal state monitoring method.

[0087] The advantages and positive effects of the present application are:

[0088] By adopting the above technical solution, the present application has the following technical effects:

[0089] The application can not only identify common abnormal types such as high temperature and high voltage, but also identify previously undefined abnormalities based on the data, by analyzing the charging history big data, inputting data containing voltage, temperature, current, soc and other information during the whole charging process.

[0090] The application can accurately determine the abnormality by taking the difference information of the battery as one of the influencing factors, because the voltage, temperature, current and soc of the battery with different degrees of new and old will have different change patterns and correlation patterns during the charging process. BRIEF DESCRIPTION OF DRAWINGS

[0091] Fig. 1 The flowchart of the preferred embodiment of the application is shown in the figure.

[0092] Fig. 2 The schematic diagram of the number of single-point fluctuations of the current curve in the preferred embodiment of the application is shown in the figure.

[0093] Fig. 3 The schematic diagram of the number of step fluctuations of the current curve in the preferred embodiment of the application is shown in the figure. DETAILED DESCRIPTION

[0094] In order to further understand the content, characteristics and effects of the application, the following embodiments are exemplified and described in detail as follows with reference to the accompanying drawings.

[0095] The following will be described in detail with reference to the accompanying drawings. Figs. 1 to 3 The technical solutions of the application will be described in detail.

[0096] A battery charging abnormal state monitoring method based on machine learning, comprising the following steps:

[0097] 1. Collecting all battery historical charging data of the battery swap station, specifically as follows:

[0098] 1.1 Data field: battery box code, temperature, voltage, current, soc, start charging time, data timestamp during the charging process;

[0099] 1.2 Data frequency: one data every 10 seconds (including all the above fields)

[0100] 2. Data preprocessing, specifically as follows:

[0101] 2.1 Group the data by battery box code and start charging time, each group representing all data of a battery during one charging process;

[0102] 2.2 Sort each group of historical charging data from low to high according to the timestamp;

[0103] 2.3 Filter out the data set with initial soc less than 40 and final soc greater than 98;

[0104] 2.4 For each set of historical charging data, filter out the data rows with soc from 40 to 98.

[0105] 3. For each set of historical charging data after pre-processing, construct the following features:

[0106] 3.1 Standard deviation of current value

[0107] 3.2 Number of single-point fluctuations in current curve

[0108] 3.3 Number of step fluctuations in current curve

[0109] 3.4 N current values obtained by interpolation method

[0110] 3.5 N voltage values obtained by interpolation method

[0111] 3.6 Maximum temperature value

[0112] 3.7 Minimum temperature value

[0113] 3.8 Average temperature value

[0114] Through the construction of features, each set of charging history data is converted into a 1 x (2N+6) feature row. The feature set is obtained, m is the number of data sets after data pre-processing.

[0115] 4. Data normalization processing is performed on the feature set, and all numerical values are scaled to the range [0, 1];

[0116] 5. PCA algorithm dimension reduction reduces the feature set from 2N+6 to K dimensions;

[0117] KNN algorithm fits the feature set after dimension reduction to obtain cluster center points, and calculates the Euclidean distance from each point to the cluster center point. The greater the distance value, the higher the abnormality. Among them:

[0118] In steps 1 and 2: an example of each set of historical charging data and the pre-processing process is as follows:

[0119] 1.1 After grouping the original data set by battery box code and starting charging time and sorting by data timestamp, an example of a set of data is as follows:

[0120] Table 1 is the original data sampling table

[0121]

[0122] 1.2 After cutting the data with soc from 40 to 100, an example is as follows:

[0123] Table 2 is the table after screening

[0124]

[0125] 2. After the data based on 1.1, 1.2, the amount of historical charging data of each group is represented as n, and the following features are constructed:

[0126] 2.1 Current standard deviation I σ

[0127] Take the current sequence I of this group of data n , and the number of this group of data is n, then the average current is:

[0128]

[0129] The current standard deviation is:

[0130]

[0131] 2.2 Number of single-point fluctuations n1 of current curve;

[0132] The current value of the battery at a certain time is I i , the current value at the previous time is I i-1 , and the current value at the next time is I i+1 , the threshold value T is set to 5A, if (I i-1 -I i >T)&(I i+1 -I i >T), I i is determined as a single-point fluctuation of the current curve, and the count is 1. For each group of historical charging data, the total number is calculated, which is n1. As shown in the current variation diagram between Fig. 2 , the number of single-point fluctuations is 3.

[0133] 2.3 Number of step fluctuations n2 of current curve

[0134] The current value of the battery at a certain time is I i , the current value at the previous time is I i-1 , and the current value at the next time is I i+1 , the threshold value T is set to 5A, if (|I i-1 -I i |≤T)&(|I i+1 -I i |≤T), I i is determined as a continuous change point.

[0135] After removing the single-point fluctuation points, remove the continuous change points, and if

[0136] (I i -Ii+1 >T)&(|I i+1 -I i+2 |≤T)&(I i+3 -I i+2 >T);

[0137] Then I i If the current curve is identified as a stepped fluctuation, it is counted as 1. For each set of historical charging data, the total number is calculated and counted as n². For example... Fig. 3 The current variation over time shown in the figure has 3 steps.

[0138] 2.4 Interpolated Current Sequence I′ N Interpolated voltage sequence U′ N ;

[0139] Take the current sequence I of each group of historical charging data n Voltage sequence U n Data timestamp sequence Time n First, calculate the difference between each time point and the starting time point to obtain the second count time series:

[0140] t i =Time i -Time1(1≤i≤n,i∈Z);

[0141] With the number of interpolations N set to 300, calculate the interpolated time series:

[0142]

[0143] The interpolated current sequence is then:

[0144]

[0145] The interpolated voltage sequence is then:

[0146]

[0147] 2.5 highest temperature T max Minimum T min Average value T mean ;

[0148] Take the temperature sequence T of the data i (1≤i≤n, i∈Z), the highest value is:

[0149] T max =max{T i}(1≤i≤n,i∈Z);

[0150] The minimum value is:

[0151] T min= min{T i}(1≤i≤n,i∈Z);

[0152] The average value is:

[0153]

[0154] 3. Perform feature construction of 2.1-2.5 on each set of historical charging data, finally obtain a feature set of (m x 606), m is the number of data sets after preprocessing of the original data, 606 is the number of features constructed for each set of historical charging data. For each column f i The following normalization processing is performed:

[0155] f max = max{f i}(1≤i≤m,i∈Z);

[0156] f min = min{f i}(1≤i≤m,i∈Z);

[0157] f i = (f i -f min ) / (f max -f min )(1≤i≤m,i∈Z);

[0158] Obtain the final feature set F(m x 606);

[0159] 4. After dimensionality reduction of the feature set using the PCA algorithm, fit the data after dimensionality reduction using the KNN algorithm to obtain the cluster center point, and calculate the Euclidean distance between each feature point X i (1≤i≤606,i∈Z) and its corresponding cluster center point C i (1≤i≤606,i∈Z):

[0160]

[0161] Arrange all feature points in descending order of L, and define the first 0.5% as abnormal feature points, and the corresponding charging group is the abnormal charging group.

[0162] A battery charging abnormal state monitoring system based on machine learning, comprising:

[0163] A basic data collection module collects historical charging data of the battery: specifically, according to the data sampling frequency, extract M rows of data fields, each row of data field includes battery box code, temperature, voltage, current, soc, start charging time, data timestamp; M is a natural number greater than 1;

[0164] Preprocessing module: preprocessing historical charging data, specifically:

[0165] S2.1, through battery box coding, starting charging time, data grouping, each group of historical charging data representing all data of a piece of battery once charging process;

[0166] S2.2, each group of historical charging data is sorted from low to high according to timestamp;

[0167] S2.3, screening out data groups with starting soc less than 40 and ending soc greater than 98;

[0168] S2.4, for each group of historical charging data, screening out data rows in the period from 40 to 98;

[0169] Construction module: construct feature set; Specifically:

[0170] S3.1, for each group of historical charging data, construct the following features: current value standard deviation, current curve single-point fluctuation number, current curve step fluctuation number, interpolated current sequence I' N , interpolated voltage sequence U' N , maximum temperature, minimum temperature and average temperature;

[0171] S3.2, through the construction of features, each group of charging historical data is converted into a 1x(2N+6) feature row; get m×(2N+6) feature set, m is the number of data groups after data preprocessing;

[0172] Normalization processing module: normalize the feature set, scale all numerical values to the range of [0, 1];

[0173] Dimension reduction module: use PCA algorithm for dimension reduction, reduce the feature set from 2N+6 to K dimensions;

[0174] Model construction module: use KNN algorithm to fit the dimension-reduced feature set to obtain the trained clustering model and clustering center point;

[0175] Abnormal analysis module: input the dimension-reduced feature set into the clustering model to output the Euclidean distance of each feature point to the clustering center point. The greater the Euclidean distance value, the higher the abnormality of the feature point.

[0176] Further: the data sampling frequency is 6 times per minute.

[0177] Further: in S3.1,

[0178] The formula for obtaining the current value standard deviation is:

[0179] Take the current sequence I nThe average current I μ is:

[0180]

[0181] The standard deviation of current I σ is:

[0182]

[0183] The current value of the battery at a certain time is I i , the current value at the previous time is I i-1 , and the current value at the next time is I i+1 , and the set threshold is T.

[0184] The acquisition process of the number of single-point fluctuations of the current curve is: if (I i-1 -I i >T)&(I i+1 -I i >T), then I i is determined as a single-point fluctuation of the current curve, and the count is 1. For each set of historical charging data, the total number is calculated, which is n1.

[0185] The acquisition process of the number of step fluctuations of the current curve is: if (|I i-1 -I i |≤T)&(|I i+1 -I i |≤T), then I i is determined as a continuous change point. After removing the single-point fluctuation points, remove the continuous change points. If

[0186] (I i -I i+1 >T)&(|I i+1 -I i+2 |≤T)&(I i+3 -I i+2 >T), then I i is determined as a step fluctuation of the current curve, and the count is 1. For each set of historical charging data, the total number is calculated, which is n2.

[0187] The interpolated current sequence I' N , the interpolated voltage sequence U' N Take the current sequence I n , the voltage sequence U n , and the data timestamp sequence Time n of each set of historical charging data. First, calculate the difference between each time point and the starting time point to get the second count time sequence:

[0188] t i = Timei -Time1(1≤i≤n,i∈z);

[0189] Take the interpolation number N, and find the interpolation time sequence:

[0190]

[0191] Then the interpolation current sequence is:

[0192]

[0193] Then the interpolation voltage sequence is:

[0194]

[0195] Further: in S3.2: perform S3.1 on each set of historical charging data, finally obtain a feature set of m x (2N+6), m is the number of data sets after preprocessing of the original data, (2N+6) is the number of features constructed by each set of historical charging data; for each column f i Perform the following normalization processing:

[0196] f max = max{f i}(1≤i≤m,i∈Z);

[0197] f min = min{f i}(1≤i≤m,i∈Z);

[0198] f i = (f i -f min ) / (f max -f min )(1≤i≤m,i∈Z);

[0199] Obtain the final feature set F.

[0200] A computer program for implementing the battery charging abnormal state monitoring method based on machine learning in the preferred embodiment described above.

[0201] An information data processing terminal for implementing the battery charging abnormal state monitoring method based on machine learning in the preferred embodiment described above.

[0202] A computer readable storage medium comprising instructions which, when executed on a computer, cause the computer to perform the battery charging abnormal state monitoring method based on machine learning in the preferred embodiment described above.

[0203] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When all or part of the embodiments are implemented in the form of a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through a wired (such as coaxial cable, optical fiber, digital subscriber line (DSL) or wireless (such as infrared, wireless, microwave, etc.)) way. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media sets. The available media can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)) and the like.

[0204] The above description is only a preferred embodiment of the present application, and does not limit the present application in any form. Any simple modification, equivalent change and modification made on the basis of the technical essence of the present application to the above embodiments are within the scope of the technical solution of the present application.

Claims

1. A machine learning based battery charging abnormal state monitoring method, characterized by, Comprise the following steps: S1: collect the historical charging data of the battery: specifically: according to the data sampling frequency, extract M rows of data fields, each row of data field includes battery box code, temperature, voltage, current, soc, start charging time, data timestamp; M is a natural number greater than 1; S2, preprocessing the historical charging data, specifically: S2.1, through the battery box code, the start charging time, data grouping, each group of historical charging data represents all data of a battery charging process; S2.2, sort each group of historical charging data from low to high according to the timestamp; S2.3, filter out the data group whose starting soc is less than 40 and ending soc is greater than 98; S2.4, each group of historical charging data, filter out the data row in the time period from 40 to 98; S3, construct the feature set; Specifically: S3.1, for each set of historical charging data, construct the following features: current value standard deviation, number of single-point fluctuations in current curve, number of step fluctuations in current curve, interpolated current sequence , interpolated voltage sequence , maximum temperature value, minimum temperature value, and average temperature value; wherein: The current value of the battery charging at a certain moment is , the current value at the previous moment is , the current value at the next moment is , and the set threshold value is T. The acquisition process of the number of single-point fluctuations of the current curve is: if , then is determined as a single-point fluctuation of the current curve, and the count is 1. For each set of historical charging data, the total number is calculated, and the count is ; The acquisition process of the number of current curve step fluctuations is: if , then determine the continuous change point; after removing the single-point fluctuation point, remove the continuous change point, and if If If the current curve is determined to be a step fluctuation, the count is 1, and the total number of each set of historical charging data is calculated as n2. S3.2, through the construction of features, each group of charging historical data is converted into a 1x(2N+6) feature row, and an m x (2N+6) feature set is obtained, m is the number of data groups after data preprocessing; S4, the feature set is normalized, and all numerical values are scaled to the range of [0, 1]; S5, dimension reduction is realized by using PCA algorithm, and the feature set is reduced from 2N+6 to K dimension; S6, the KNN algorithm is used to fit the feature set after dimension reduction, and the trained clustering model and clustering center point are obtained; S7, input the feature set after dimension reduction into the clustering model, output the euclidean distance from each feature point to the clustering center point, the greater the euclidean distance value, the higher the abnormality of the feature point.

2. The machine learning based battery charge abnormal state monitoring method of claim 1, wherein: The data sampling frequency is 6 times per minute.

3. The machine learning based battery charge abnormal state monitoring method of claim 1, wherein: In S3.1: The formula for obtaining the current value standard deviation is: Taking the current sequence in the present set of data The number of data in the present set is n, and the average value of the current is ​ ； Current standard deviation Is: ； interpolated current sequence interpolated voltage sequence current sequence of each set of historical charging data voltage sequence data timestamp sequence first, the difference between each time point and the starting time point is calculated to obtain a second count time sequence: ； Take the interpolation number N, and calculate the interpolation time sequence: ； Then the interpolation current sequence is: ； Then the interpolation voltage sequence is: 。 4. The machine learning based battery charge abnormal state monitoring method of claim 3, wherein: In S3.2: S3.1 is performed on each set of historical charging data, finally obtaining a feature set of m x (2N+6), m is the number of data sets after preprocessing of the original data, and 2N+6 is the number of features constructed from each set of historical charging data; each column of the feature set is normalized as follows: ​ ； ； ； Get the final feature set F.

5. A machine learning based battery charging abnormal state monitoring system, characterized by, Comprise: The basic data collection module: collect the historical charging data of the battery: specifically: according to the data sampling frequency, extract M rows of data fields, each row of data field includes battery box code, temperature, voltage, current, soc, start charging time, data timestamp; M is a natural number greater than 1; The preprocessing module: preprocessing the historical charging data, specifically: S2.1, through the battery box code, the start charging time, data grouping, each group of historical charging data represents all data of a battery charging process; S2.2, sort each group of historical charging data from low to high according to the timestamp; S2.3, filter out the data group whose starting soc is less than 40 and ending soc is greater than 98; S2.4, each group of historical charging data, filter out the data row in the time period from 40 to 98; The construction module: construct the feature set; Specifically: S3.1, for each set of historical charging data, construct the following features: current value standard deviation, number of single-point fluctuations in current curve, number of step fluctuations in current curve, interpolated current sequence , interpolated voltage sequence , maximum temperature value, minimum temperature value, and average temperature value; wherein: The current value of the battery charging at a certain moment is , the current value at the previous moment is , the current value at the next moment is , and the set threshold value is T. The acquisition process of the number of single-point fluctuations of the current curve is: if , then is determined as a single-point fluctuation of the current curve, the count is 1, the total number is calculated for each set of historical charging data, and the count is ; The acquisition process of the number of current curve step fluctuations is: if , then determine the continuous change point; after removing the single-point fluctuation point, remove the continuous change point, and if the remaining points are If If the current curve is determined to be a step fluctuation, the count is 1, and the total number of each set of historical charging data is calculated as n2. S3.2, through the construction of features, each group of charging historical data is converted into a 1x(2N+6) feature row; Get m x (2N+6) feature set, m is the number of data groups after data preprocessing; The normalization processing module: the feature set is normalized, and all numerical values are scaled to the range of [0, 1]; The dimension reduction module: dimension reduction is realized by using PCA algorithm, and the feature set is reduced from 2N+6 to K dimension; Model construction module: using KNN algorithm to fit the feature set after dimensionality reduction, to obtain the trained clustering model and clustering center point; Abnormal analysis module: input the feature set after dimensionality reduction into the clustering model, output the euclidean distance from each feature point to the clustering center point, the greater the euclidean distance value, the higher the abnormality of the feature point.

6. The machine learning based battery charge abnormal state monitoring system of claim 5, wherein: The data sampling frequency is 6 times per minute.

7. The machine learning based battery charge abnormal state monitoring system of claim 5, wherein: In S3.1: The formula for obtaining the current value standard deviation is: Taking the current sequence in the present set of data The average value of the current is n, then the average value of the current is ； Current standard deviation Is: ； interpolated current sequence interpolated voltage sequence current sequence of each set of historical charging data voltage sequence data timestamp sequence first, the difference between each time point and the start time point is calculated to obtain a second count time sequence: ； Take the interpolation number N, and calculate the interpolation time sequence: ； Then the interpolation current sequence is: ； Then the interpolation voltage sequence is: 。 8. The machine learning based battery charge abnormal state monitoring system of claim 7, wherein: In S3.2: S3.1 is performed on each set of historical charging data, finally obtaining a feature set of m x (2N+6), m is the number of data sets after preprocessing of the original data, (2N+6) is the number of features constructed for each set of historical charging data; the following normalization processing is performed on each column of the feature set: ​ ； ； ； Get the final feature set F.

9. An information data processing terminal of a battery charge abnormal state monitoring method, characterized by: A computer program product for implementing the machine learning-based battery charging abnormal state monitoring method of any one of claims 1-4.

10. A computer-readable storage medium, characterized in that: A computer program product for implementing the machine learning-based battery charging abnormal state monitoring method of any one of claims 1-4.

Images