Lithium battery charging data monitoring method and system

Abstract

The invention relates to the field of data processing, in particular to a lithium battery charging data monitoring method and system. The method comprises the steps of obtaining feature sequences of historical charging data points, obtaining feature data of any dimension from the feature sequences of all the historical charging data points, calculating category consistency between the feature data of each dimension and the feature data of other dimensions, and determining a reference weight of the feature data of each dimension according to the category consistency; obtaining classification meticulousness of the feature data of each dimension, and obtaining a time sequence weight adjustment coefficient of the feature data of each dimension according to the classification meticulousness; and setting the weight of each clustering stage according to the time sequence weight adjustment coefficient and the reference weight, and enabling the weight of each clustering stage to participate in clustering iteration updating to complete clustering processing so as to realize anomaly detection of the charging data point at the current moment. The accuracy of charging detection is improved by improving the accuracy of clustering.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a method and system for monitoring lithium battery charging data. Background Art

[0002] With the rapid development of new energy technologies, lithium-ion batteries, with their advantages such as high energy density and long cycle life, are widely used in electric vehicles, energy storage systems, and portable electronic devices. The safe and efficient use of lithium-ion batteries requires precise monitoring of the charging process. Promptly identifying abnormal charging data points and taking appropriate action is crucial for extending battery life and ensuring safe operation of equipment.

[0003] Currently, cluster analysis is a common technical means of identifying abnormal data in the monitoring of lithium battery charging data points. Traditional clustering methods usually adopt a single clustering strategy when processing lithium battery charging data points, or only refer to charging data points of coarse classification dimensions, or only rely on charging data points of fine classification dimensions. When only referring to data of coarse classification dimensions, although macro clustering can be achieved quickly, due to insufficient data feature information, it is difficult to accurately distinguish subtle differences, resulting in rough clustering results and inability to effectively identify potential abnormal charging data points; while relying solely on data of fine classification dimensions for clustering can achieve detailed clustering, due to the high data dimension and high computational complexity, it is very easy to fall into local optimal solutions during the clustering process, and it is impossible to obtain the globally optimal clustering results, which reduces the accuracy and reliability of abnormal data monitoring.

[0004] Furthermore, existing clustering analysis methods for lithium battery charging data points lack targeted processing of data characteristics at different stages of the clustering process. This fails to fully leverage the advantages of data from different dimensions in clustering, resulting in poor overall clustering results and difficulty meeting the growing demand for safe and efficient lithium battery monitoring. Therefore, a method is urgently needed that can improve clustering accuracy, effectively prevent local optima, and achieve comprehensive and accurate monitoring of lithium battery charging data points. Summary of the Invention

[0005] In order to solve the problem of how to prevent falling into local optimum and achieve accurate classification, the present invention provides a lithium battery charging data monitoring method and system.

[0006] In a first aspect, the present invention provides a method for monitoring charging data of a lithium battery, which adopts the following technical solution: A method for monitoring charging data of a lithium battery comprises the following steps: Obtain the current charging data point of the lithium battery and the corresponding historical charging data point; Obtaining a feature sequence of historical charging data points, wherein the feature data in the feature sequence reflects the values ​​and change trends of the historical charging data points; Acquire feature data of any dimension from the feature sequence of all historical charging data points, calculate the category consistency of the feature data of each dimension with the feature data of other dimensions, and determine the baseline weight of the feature data of each dimension based on the category consistency, wherein the baseline weight is positively correlated with the category consistency; obtain the classification meticulousness of the feature data of each dimension, wherein the classification meticulousness represents the refinement of the classification, and obtain the time series weight adjustment coefficient of the feature data of each dimension based on the classification meticulousness, wherein the weight adjustment coefficient is positively correlated with the classification meticulousness; The weight of each clustering stage is set according to the time series weight adjustment coefficient and the benchmark weight, and the weight of each clustering stage is involved in the clustering iterative update to complete the clustering process to achieve anomaly detection of charging data points at the current moment.

[0007] The present invention introduces a feature sequence to reflect the information on the value and change trend of the charging data point, thereby preventing the situation where the abnormality of lithium battery charging cannot be accurately reflected by only using the value; further, by analyzing the category consistency, the contribution of each dimensional data to the clustering is reflected, thereby giving a larger weight to the data with a large contribution degree, thereby improving the accuracy of clustering; further, based on the classification detail, a time series weight adjustment coefficient is set to adjust the weight of each stage, thereby making the data of the dimension with small classification detail have a larger weight in the early stage of clustering, thereby realizing macro classification in the early stage of clustering; making the data of the dimension with large classification detail have a larger weight in the late stage of clustering, thereby realizing more detailed classification in the late stage of clustering, preventing falling into local optimality, and improving the accuracy of classification.

[0008] Preferably, the step of obtaining a characteristic sequence of historical charging data points includes: Obtaining data of each dimension in the historical charging data point, and obtaining a historical charging data sequence corresponding to the data of each dimension in the historical charging data point; The historical charging data sequence is processed using the Gaussian pyramid algorithm to obtain several layers of processed data sequences. In each layer of processed data sequence, the data corresponding to each dimension of historical charging data in each historical charging data point is obtained and recorded as the macro data of each dimension of data in the historical charging data point; the sequence consisting of all dimensional data in the historical charging data point, the slopes of all dimensional data, and the slopes of all macro data of all historical charging data points is recorded as a feature sequence.

[0009] The present invention uses Gaussian pyramid to process historical charging data sequences to obtain trend information of different macro situations, thereby more comprehensively extracting change information of charging data and providing an information basis for accurate anomaly detection.

[0010] Preferably, calculating the category consistency of the feature data of each dimension and the feature data of other dimensions includes: Performing statistics on the feature data of each dimension to obtain a statistical histogram of the feature data of each dimension; Obtaining a characteristic sequence corresponding to characteristic data in each independent peak in the statistical histogram and recording it as the analysis characteristic sequence of each independent peak; Calculate the distribution consistency of each independent peak and other dimensions; The normalized value of the mean of the distribution consistency of all independent peaks of the statistical histogram of the feature data of each dimension is used as the category consistency between the feature data of each dimension and the feature data of other dimensions.

[0011] The present invention accurately reflects the consistency of data classification in different dimensions by introducing the distribution consistency of independent peaks and other dimensions.

[0012] Preferably, the calculation of the distribution consistency of each independent peak with other dimensions includes:

[0013] in, The probability that the analysis feature sequence of each independent peak is distributed in the j-th independent peak in the statistical histogram of the feature data of the i-th dimension is, represents the logarithmic function with base 10, Represents the number of independent peaks in the statistical histogram of the feature data of the i-th dimension, represents the number of dimensions of the feature sequence, Indicates the distribution consistency of each independent peak with other dimensions.

[0014] The present invention reflects the centralized consistency of independent peaks and data in other dimensions by analyzing the information entropy of the distribution of feature sequences corresponding to feature data in independent peaks in other dimensions, thereby providing a basis for accurately analyzing the classification consistency of different dimensions.

[0015] Preferably, the step of determining the reference weight of the feature data of each dimension according to category consistency includes: The ratio of the category consistency to the cumulative sum of the category consistency of all dimensions is used as the benchmark weight of the feature data of each dimension.

[0016] Preferably, obtaining the classification detail of the feature data of each dimension includes: The normalized value of the number of independent peaks in the statistical histogram of the feature data of each dimension is used as the classification fineness of the feature data of each dimension.

[0017] Preferably, the step of obtaining the temporal weight adjustment coefficient of the feature data of each dimension according to the classification detail includes: The product of the classification detail and the preset adjustment coefficient is used as the temporal weight adjustment coefficient of the feature data of each dimension.

[0018] Preferably, the weight of each clustering stage is set according to the time series weight adjustment coefficient and the reference weight, and the weight of each clustering stage is involved in the clustering iterative update to complete the clustering process, including: The normalized value of the ratio of the baseline weight to the timing weight adjustment coefficient is used as the weight of the first clustering stage. If the weight of the first clustering stage is less than the preset weight threshold, the feature data of the dimension is removed to obtain the adjusted feature sequence; the weight of the first clustering stage is used in the clustering iterative update; in response to the stability of the clustering being greater than the preset stability threshold, the normalized value of the product of the baseline weight and the timing weight adjustment coefficient is used as the weight of the second clustering stage in the clustering iterative update, until the clustering cutoff condition is met and the clustering is completed.

[0019] The present invention uses the time series weight adjustment coefficient to adjust the weights of data of different dimensions in each clustering stage, so that the data of dimensions with low classification detail have a larger weight in the early stage of clustering, thereby increasing the ability to grasp the macro in the early stage of clustering and preventing falling into local optimality; the data of dimensions with high classification detail have a smaller weight in the late stage of clustering, thereby increasing the ability to grasp detailed information in the late stage of clustering and improving the accuracy of classification.

[0020] Preferably, the method for realizing abnormal detection of charging data points at the current moment includes: Obtain the feature sequence of the charging data point at the current moment, input the feature sequence of the charging data point at the current moment into the clustering result to obtain the category to which the charging data point at the current moment belongs, and judge the abnormal situation based on the distance between the charging data point at the current moment and the cluster center of the category to which it belongs.

[0021] In a second aspect, the present invention provides a lithium battery charging data monitoring system, which adopts the following technical solution: A lithium battery charging data monitoring system includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned lithium battery charging data monitoring method is implemented.

[0022] By adopting the above technical solution, the above-mentioned lithium battery charging data monitoring method is generated into a computer program and stored in a memory so as to be loaded and executed by a processor, thereby making a terminal device based on the memory and the processor for easy use.

[0023] The present invention has the following technical effects: The present invention introduces a characteristic sequence to reflect the value and change trend of the charging data point, thereby preventing the situation where only the value cannot accurately reflect the abnormality of lithium battery charging; Furthermore, by analyzing the category consistency to reflect the contribution of each dimension of data to the clustering, a larger weight is set for the data with greater contribution, thereby improving the accuracy of clustering; Furthermore, based on the classification detail, the time series weight adjustment coefficient is set to adjust the weight of each stage, so that the data of the dimension with small classification detail has a larger weight in the early stage of clustering, thereby achieving more grasp of macro information in the early stage of clustering and preventing falling into local optimality; the data of the dimension with large classification detail has a larger weight in the late stage of clustering, thereby achieving more grasp of detailed information in the late stage of clustering and improving the accuracy of classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 The present invention is a flowchart of a method for monitoring charging data of a lithium battery according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The embodiment of the present invention discloses a method for monitoring charging data of a lithium battery, referring to Figure 1 , including steps S1 to S4: S1: Obtain the current charging data point of the lithium battery and the corresponding historical charging data point.

[0026] Specifically, each type of charging data of the lithium battery at the current moment is obtained, and all types of charging data of the lithium battery at the current moment constitute a data point and are recorded as a charging data point.

[0027] The charging data points at the historical moments corresponding to the current moment are obtained and recorded as historical charging data points.

[0028] The historical moment corresponding to the current moment is a historical moment in the same charging stage as the current moment.

[0029] The types of charging data may be charging voltage, charging current, charging temperature, state of charge, and electrode material impedance, or other types, which are not specifically limited in this embodiment.

[0030] S2: Acquire a feature sequence of historical charging data points, where the feature data in the feature sequence reflects the values ​​and change trends of the historical charging data points.

[0031] It should be noted that charging anomalies can sometimes manifest as abnormal charging data, and sometimes as abnormal charging data trends. Therefore, to accurately reflect the charging status of lithium batteries, both the values ​​and trends must be considered. Therefore, before cluster analysis, a characteristic indicator must be constructed to reflect both the values ​​and trends of charging data.

[0032] Preferably, as an example, obtaining a feature sequence of historical charging data points includes: Obtaining data of each dimension in the historical charging data point, and obtaining a historical charging data sequence corresponding to the data of each dimension in the historical charging data point; The historical charging data sequence is processed using the Gaussian pyramid algorithm to obtain several layers of processed data sequences. In each layer of processed data sequence, the data corresponding to each dimension of historical charging data at each historical charging data point is obtained and recorded as the macro data of each dimension of the historical charging data point. The sequence consisting of all dimensional data in the historical charging data point, the slopes of all dimensional data, and the slopes of all macro data of all historical charging data points is recorded as the feature sequence. It is understandable that since the collected data is affected by factors such as noise, only using one level of change information cannot accurately reflect the trend of charging data. Therefore, Gaussian pyramid processing can reflect trend information at different levels.

[0033] It should be added that the method for obtaining the historical charging data series includes: Obtain any dimension data in the historical data point, obtain the time series data of the entire process in which the dimension data is located, and arrange the obtained time series data in time sequence to obtain a historical charging data sequence.

[0034] It should be further supplemented that the methods for obtaining the slopes of the various dimensional data and the slopes of the various macro data include: The difference between each dimension data and the previous data in the corresponding historical charging data sequence is used as the slope of each dimension data; The difference between each macro data and the previous data in the processed data sequence of the corresponding layer is taken as the slope of each macro data.

[0035] S3: Obtain feature data of any dimension in the feature sequence of all historical charging data points, calculate the category consistency of the feature data of each dimension with the feature data of other dimensions, determine the baseline weight of the feature data of each dimension based on the category consistency, and the baseline weight is positively correlated with the category consistency; obtain the classification detail of the feature data of each dimension, and the classification detail characterizes the fineness of the classification, and obtain the time series weight adjustment coefficient of the feature data of each dimension based on the classification detail, and the weight adjustment coefficient is positively correlated with the classification detail.

[0036] It should be noted that in order to avoid falling into local optimality and achieve accurate classification analysis, it is necessary to make the data of the macro-level classification dimension play a greater role in the early stage of clustering, so as to better grasp the macro information of the data in the early stage of clustering and complete the macro clustering analysis. In order to prevent the failure of completing more detailed classification by clustering analysis at the macro level, it is necessary to find the cluster analysis at the macro level and then make use of the data of the more detailed analysis dimension to play a greater role, so as to conduct detailed classification analysis based on the macro analysis.

[0037] S30: Acquire feature data of any dimension from the feature sequence of all historical charging data points, calculate the category consistency of the feature data of each dimension with the feature data of other dimensions, and determine the benchmark weight of the feature data of each dimension based on the category consistency.

[0038] It should be noted that because each dimension of the characteristic sequence of historical charging data points contributes differently to classification, some dimensions can facilitate classification, while others can interfere with it. To achieve accurate classification, the contribution of each dimension's data must be considered. Therefore, before analyzing the macro and micro aspects of each dimension's classification, it's important to first analyze its contribution and assign a baseline weight to each dimension based on its contribution.

[0039] It should be further explained that the dimensions that can promote classification should have classification consistency with other dimensions. If the classification consistency of a dimension is low with other dimensions, it means that the data of this dimension is inaccurate or the data of this dimension cannot promote classification analysis. Therefore, the benchmark weight can be set based on the classification consistency.

[0040] Preferably, as an example, feature data of any dimension is obtained from the feature sequence of all historical charging data points, the category consistency of the feature data of each dimension and the feature data of other dimensions is calculated, and the benchmark weight of the feature data of each dimension is determined according to the category consistency, including: Obtain feature data of any dimension from the feature sequence of all historical charging data points, and perform statistics on the feature data of each dimension to obtain a statistical histogram of the feature data of each dimension; The statistical histogram is filtered to obtain the minimum value points in the statistical histogram after filtering, and the area between each two minimum value points of the statistical histogram is regarded as an independent peak.

[0041] In particular, if there is no minimum point before the minimum point, the area between the first point and the extreme point is taken as an independent peak. If there is no minimum point after the minimum point, the area between the minimum point and the last point is taken as an independent peak.

[0042] Obtaining a characteristic sequence corresponding to characteristic data in each independent peak in the statistical histogram and recording it as the analysis characteristic sequence of each independent peak; Calculate the distribution consistency of each independent peak with other dimensions:

[0043] in, The probability that the analysis feature sequence of each independent peak is distributed in the j-th independent peak in the statistical histogram of the feature data of the i-th dimension is, represents the logarithmic function with base 10, Represents the number of independent peaks in the statistical histogram of the feature data of the i-th dimension, represents the number of dimensions of the feature sequence, Indicates the distribution consistency of each independent peak with other dimensions.

[0044] It is understandable that, under normal circumstances, if the data classification of one dimension is consistent with that of other dimensions, the aggregation characteristics of the data on this dimension should also be reflected in other dimensions. For example, if the data on one dimension shows the aggregation characteristics of an independent peak, even if it does not show the aggregation characteristics of an independent peak in other dimensions, it should be concentrated in a small number of independent peaks, rather than being evenly distributed in all independent peaks. This aggregation feature should be in order to meet the performance requirements of classification consistency. It reflects the discrete distribution of the analysis feature sequence of the independent peak in other dimensions. The larger the value, the more evenly distributed all the independent peaks in the analysis feature sequence of the independent peak are in other dimensions. That is to say, the aggregation characteristics of the independent peak are not reflected in other dimensions, and thus the classification consistency between the independent peak and other dimensions is poor.

[0045] The normalized value of the mean of the distribution consistency of all independent peaks of the statistical histogram of the feature data of each dimension is used as the category consistency between the feature data of each dimension and the feature data of other dimensions.

[0046] The ratio of the category consistency to the cumulative sum of the category consistency of all dimensions is used as the benchmark weight of the feature data of each dimension.

[0047] It can be understood that the greater the category consistency, the greater the classification consistency of the data in this dimension and the data in other dimensions. Therefore, the data in this dimension is easier to promote classification, and therefore the weight of the data in this dimension should be increased.

[0048] It should be added that the method for obtaining the probability that the analysis feature sequence of each independent peak is distributed in the j-th independent peak in the statistical histogram of the feature data in the i-th dimension includes: Obtain the number of analysis feature sequences of each independent peak, obtain the number of j-th independent peaks in the statistical histogram of the feature data of the corresponding dimension in the analysis feature sequence of each independent peak, record it as the distribution number of the j-th independent peak, divide the distribution number of the j-th independent peak by the number of analysis feature sequences of each independent peak, and obtain the probability that the analysis feature sequence of each independent peak is distributed in the j-th independent peak in the statistical histogram of the feature data of the i-th dimension.

[0049] S31: Obtaining classification detail of feature data of each dimension, where the classification detail represents the fineness of the classification, and obtaining a temporal weight adjustment coefficient of the feature data of each dimension according to the classification detail.

[0050] It should be noted that in order to pay more attention to the data information of the dimensions with higher macro classification in the early stage of clustering, and pay more attention to the data information of the dimensions with higher detailed classification in the later stage of clustering, it is necessary to analyze the detailed classification of the data in each dimension.

[0051] It should be further explained that the values ​​of the same type of data in one dimension are more similar, and the values ​​of the same type of data in one dimension have clustering characteristics. Therefore, the more detailed the data classification in one dimension is, the more independent peaks there are in the data in one dimension.

[0052] Preferably, as an example, obtaining the classification detail of the feature data of each dimension, and obtaining the time series weight adjustment coefficient of the feature data of each dimension according to the classification detail, includes: The normalized value of the number of independent peaks in the statistical histogram of the feature data of each dimension is used as the classification fineness of the feature data of each dimension.

[0053] The product of the classification detail and the preset adjustment coefficient is used as the temporal weight adjustment coefficient of the feature data of each dimension.

[0054] It can be understood that the number of independent peaks can reflect the number of categories of data on this dimension. The more categories there are, the more detailed the data classification on this dimension.

[0055] S4: The weight of each clustering stage is set according to the time series weight adjustment coefficient and the benchmark weight, and the weight of each clustering stage is involved in the clustering iterative update to complete the clustering process to achieve anomaly detection of the charging data point at the current moment.

[0056] S40: setting the weight of each clustering stage according to the time series weight adjustment coefficient and the reference weight, and participating the weight of each clustering stage in clustering iterative update to complete the clustering process.

[0057] It should be noted that in order to prioritize data with high macroscopic categorization in the early stages of clustering, that is, to prioritize data with low categorization detail, the weight of this data should be assigned higher weights in the early stages. In order to prioritize data with high categorization detail in the later stages of clustering, the weight of this data should be assigned higher weights in the latter stages. To reduce the interference of data with high categorization detail and low contribution in the early stages of clustering, and to reduce computational complexity, it is advisable to remove data from these dimensions with high categorization detail and low contribution from clustering analysis.

[0058] Preferably, as an example, the weight of each clustering stage is set according to the time series weight adjustment coefficient and the reference weight, and the weight of each clustering stage is involved in the clustering iterative update to complete the clustering process, including: The normalized value of the ratio of the baseline weight to the time series weight adjustment coefficient is used as the weight of the first clustering stage. If the weight of the first clustering stage is less than the preset weight threshold, the feature data of the dimension is removed to obtain the adjusted feature sequence; clustering is iteratively updated based on the weight of the first clustering stage and the adjusted feature sequence; in response to the stability of the cluster being greater than the preset stability threshold, the normalized value of the product of the baseline weight and the time series weight adjustment coefficient is used as the weight of the second clustering stage to participate in the clustering iterative update until the clustering cutoff condition is met and the clustering is completed. This embodiment performs cluster analysis based on the K-means clustering algorithm. Other embodiments may use other clustering algorithms for cluster analysis, and this embodiment does not impose specific restrictions. The clustering cutoff condition can use the clustering cutoff condition that comes with the K-means clustering algorithm, which will not be repeated here.

[0059] It is understandable that in the early stages of clustering, by dividing the baseline weight by the time-series weight adjustment coefficient, the weights of dimensions with high classification detail are set lower, allowing more attention to be paid to the data information of dimensions with low classification detail in the early stages of clustering, thereby achieving macro-clustering in the early stages of clustering. In the later stages of clustering, by multiplying the baseline weight by the time-series weight adjustment coefficient, the weights of dimensions with high classification detail are set higher, allowing more attention to be paid to the data information of dimensions with high classification detail in the later stages of clustering, thereby achieving detailed clustering in the later stages of clustering. By removing data from dimensions with low weights, the amount of data computation can be reduced.

[0060] It should be added that the methods for obtaining clustering stability include: The cluster centers of each clustering stage are obtained, the reciprocal of the distance between the cluster center of a clustering stage and the cluster center of the previous clustering stage is taken as the stability of the cluster center, and the mean of the stabilities of all cluster centers is taken as the stability of the cluster.

[0061] It should be further added that clustering iteration update is performed based on the weights of the first clustering stage and the adjusted feature sequence, including: The weights based on the first clustering stage are used as the weights of the feature data of each dimension in the adjusted feature sequence to calculate the distance between the cluster center and each adjusted feature sequence. Based on the distance between the cluster center and each adjusted feature sequence, the cluster center closest to the adjusted feature sequence is obtained, and the adjusted feature sequence is divided into the range of the cluster center. The mean of the adjusted feature sequence within the range of the cluster center is used as the updated cluster center.

[0062] The normalized value of the product of the benchmark weight and the time series weight adjustment coefficient is used as the weight of the second clustering stage to participate in the clustering iterative update, which is the same as the above clustering iterative update process and will not be repeated here.

[0063] S41: To realize abnormality detection of charging data points at the current moment.

[0064] Preferably, as an example, to implement abnormality detection of charging data points at the current moment, the method includes: Obtain a characteristic sequence of the current charging data point, input this characteristic sequence into the clustering result to determine the category to which the current charging data point belongs, obtain the distance between the current charging data point and the cluster center of the category to which it belongs, and if the distance between the current charging data point and the cluster center of the category to which it belongs is greater than a preset distance threshold, determine that there is an abnormality in charging at the current moment. This embodiment uses the upper quartile of the distance between the characteristic sequence of all charging data points in the category and the cluster center as the preset distance threshold. Other embodiments may use other methods to set the preset distance threshold, and this embodiment does not specifically limit this.

[0065] An embodiment of the present invention further discloses a lithium battery charging data monitoring system, comprising a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, a lithium battery charging data monitoring method according to the present invention is implemented.

[0066] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and will not be described in detail here.

[0067] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as resistive random access memory, dynamic random access memory, static random access memory, enhanced dynamic random access memory, high bandwidth memory, hybrid memory cube, etc., or any other medium that can be used to store the required information and can be accessed by an application, module, or both. Any such computer storage medium may be part of, accessible to, or connectable to the device.

Claims

1. A lithium battery charging data monitoring method, characterized in that: Including steps: Obtain the current charging data point of the lithium battery and the corresponding historical charging data point; Obtaining a feature sequence of historical charging data points, wherein the feature data in the feature sequence reflects the values ​​and change trends of the historical charging data points; Acquire feature data of any dimension from the feature sequence of all historical charging data points, calculate the category consistency of the feature data of each dimension with the feature data of other dimensions, and determine the baseline weight of the feature data of each dimension based on the category consistency, wherein the baseline weight is positively correlated with the category consistency; obtain the classification meticulousness of the feature data of each dimension, wherein the classification meticulousness represents the refinement of the classification, and obtain the time series weight adjustment coefficient of the feature data of each dimension based on the classification meticulousness, wherein the weight adjustment coefficient is positively correlated with the classification meticulousness; The weight of each clustering stage is set according to the time series weight adjustment coefficient and the benchmark weight, and the weight of each clustering stage is involved in the clustering iterative update to complete the clustering process to achieve anomaly detection of charging data points at the current moment.

2. A lithium battery charging data monitoring method according to claim 1, characterized in that: The step of obtaining a characteristic sequence of historical charging data points includes: Obtaining data of each dimension in the historical charging data point, and obtaining a historical charging data sequence corresponding to the data of each dimension in the historical charging data point; The historical charging data sequence is processed using the Gaussian pyramid algorithm to obtain several layers of processed data sequences. In each layer of processed data sequence, the data corresponding to each dimension of historical charging data in each historical charging data point is obtained and recorded as the macro data of each dimension of data in the historical charging data point; the sequence consisting of all dimensional data in the historical charging data point, the slopes of all dimensional data, and the slopes of all macro data of all historical charging data points is recorded as a feature sequence.

3. The method for monitoring lithium battery charging data according to claim 1, wherein: Calculating the category consistency of feature data of each dimension and feature data of other dimensions includes: Performing statistics on the feature data of each dimension to obtain a statistical histogram of the feature data of each dimension; Obtaining a characteristic sequence corresponding to characteristic data in each independent peak in the statistical histogram and recording it as the analysis characteristic sequence of each independent peak; Calculate the distribution consistency of each independent peak and other dimensions; The normalized value of the mean of the distribution consistency of all independent peaks of the statistical histogram of the feature data of each dimension is used as the category consistency between the feature data of each dimension and the feature data of other dimensions.

4. A lithium battery charging data monitoring method according to claim 3, characterized in that: The calculation of the distribution consistency of each independent peak and other dimensions includes: in, The probability that the analysis feature sequence of each independent peak is distributed in the j-th independent peak in the statistical histogram of the feature data of the i-th dimension is, represents the logarithmic function with base 10, Represents the number of independent peaks in the statistical histogram of the feature data of the i-th dimension, represents the number of dimensions of the feature sequence, Indicates the distribution consistency of each independent peak with other dimensions.

5. The method for monitoring lithium battery charging data according to claim 1, wherein: Determining the benchmark weight of feature data of each dimension based on category consistency includes: The ratio of the category consistency to the cumulative sum of the category consistency of all dimensions is used as the benchmark weight of the feature data of each dimension.

6. A lithium battery charging data monitoring method according to claim 3, characterized in that: The classification details of the feature data of each dimension are obtained, including: The normalized value of the number of independent peaks in the statistical histogram of the feature data of each dimension is used as the classification fineness of the feature data of each dimension.

7. A lithium battery charging data monitoring method according to claim 3, characterized in that: The temporal weight adjustment coefficient of the feature data of each dimension is obtained according to the classification detail, including: The product of the classification detail and the preset adjustment coefficient is used as the temporal weight adjustment coefficient of the feature data of each dimension.

8. The method for monitoring lithium battery charging data according to claim 1, wherein: The weight of each clustering stage is set according to the time series weight adjustment coefficient and the reference weight, and the weight of each clustering stage is involved in the clustering iterative update to complete the clustering process, including: The normalized value of the ratio of the baseline weight to the timing weight adjustment coefficient is used as the weight of the first clustering stage. If the weight of the first clustering stage is less than the preset weight threshold, the feature data of the dimension is removed to obtain the adjusted feature sequence; the weight of the first clustering stage is used in the clustering iterative update; in response to the stability of the clustering being greater than the preset stability threshold, the normalized value of the product of the baseline weight and the timing weight adjustment coefficient is used as the weight of the second clustering stage in the clustering iterative update, until the clustering cutoff condition is met and the clustering is completed.

9. The method for monitoring lithium battery charging data according to claim 1, wherein: The method for realizing abnormal detection of charging data points at the current moment includes: Obtain the feature sequence of the charging data point at the current moment, input the feature sequence of the charging data point at the current moment into the clustering result to obtain the category to which the charging data point at the current moment belongs, and judge the abnormal situation based on the distance between the charging data point at the current moment and the cluster center of the category to which it belongs.

10. A lithium battery charging data monitoring system, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a lithium battery charging data monitoring method according to any one of claims 1 to 9 is implemented.

Images