Method for Monitoring Driving State of Electric Drive System of New Energy Vehicle Based on Machine Learning

Abstract

The present invention relates to the technical field of abnormal detection of new energy vehicles, and specifically relates to a method for monitoring the driving state of the electric drive system of new energy vehicles based on machine learning. The method first screens reference clustering center points according to the density of voltage and temperature data points in the neighborhood distribution and their relative positions. Further, according to the change in the density of data points on the line connecting the voltage and temperature data points and the reference clustering center points, and the slope of the connection line, the clustering influence weight is obtained. The clustering influence weight affects the clustering distance during clustering, so that abnormal voltage and temperature data points can be more likely to be assigned to abnormal voltage and temperature data clustering clusters, that is, a more accurate clustering analysis result is obtained. Thus, a more accurate clustering abnormality degree can be obtained according to the density of data points and the relative outlier situation of the clustering clusters, and the effect of monitoring the driving state of the electric drive system of new energy vehicles according to the clustering abnormality degree is better.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormal detection of new energy vehicles, and particularly to a method for monitoring the driving state of the electric drive system of new energy vehicles based on machine learning. Background Art

[0002] The electric drive system of a new energy vehicle refers to the overall system composed of an electric motor, a battery, an electronic control system, etc. used in an electric vehicle. By converting electrical energy into mechanical energy, it drives the vehicle to move, which is very important for new energy vehicles. Therefore, it is necessary to monitor the driving state of the electric drive system of new energy vehicles to timely detect abnormalities in the electric drive system and prevent impacts on vehicle driving. Considering that the electric drive system is a device that converts electrical energy into mechanical energy, there will inevitably be voltage and heat generated by electrical energy loss during the operation of the electric drive system. Therefore, when the state of the electric drive system is abnormal, the corresponding abnormalities will be reflected in the voltage data and temperature data. Therefore, the abnormal detection is usually carried out based on the motor voltage data and motor temperature data of the electric drive system of new energy vehicles, so as to monitor the driving state of the electric drive system of new energy vehicles.

[0003] The prior art usually uses the k-means clustering algorithm to perform clustering analysis on the voltage-temperature data points corresponding to the motor voltage data and motor temperature data. By analyzing the dispersion degree and outlier degree of each obtained clustering cluster, the clustering abnormality degree is analyzed, and then the driving state of the electric drive system of new energy vehicles is monitored according to the clustering abnormality degree. However, the basis for the k-means clustering algorithm to perform clustering analysis is the Euclidean distance between voltage-temperature data points. However, when an abnormality occurs in the driving of the electric drive system of new energy vehicles, the difference in numerical values between the temperature-voltage data in the abnormal situation and the voltage-temperature data in the normal situation is relatively small, and the voltage and temperature after the new energy vehicle starts change linearly, resulting in the abnormality of the abnormal voltage-temperature data points not being obvious enough, and the ability to divide the abnormal voltage-temperature data points into the abnormal data point clustering cluster is poor. That is, the abnormal voltage-temperature data points will be divided into the clustering cluster corresponding to the normal data points, causing the clustering process to fall into a local optimal solution, and the corresponding clustering result is not accurate enough. That is, the accuracy of the clustering analysis result of the voltage-temperature data points by the k-means clustering algorithm is relatively low, further resulting in a low accuracy of the clustering abnormality degree of the voltage-temperature data clustering cluster in the calculated clustering result, and making the effect of monitoring the driving state of the electric drive system of new energy vehicles according to the clustering abnormality degree poor. Summary of the Invention

[0004] To solve the technical problem that the accuracy of the clustering analysis result of voltage-temperature data points by the k-means clustering algorithm is low, resulting in a low accuracy of the clustering anomaly degree of the voltage-temperature data clustering clusters in the calculated clustering result, and making the effect of monitoring the driving state of the new energy vehicle electric drive system according to the clustering anomaly degree poor, the purpose of the present invention is to provide a method for monitoring the driving state of the new energy vehicle electric drive system based on machine learning, and the specific technical solution adopted is as follows:

[0005] The present invention proposes a method for monitoring the driving state of the new energy vehicle electric drive system based on machine learning, and the method includes:

[0006] Obtain all voltage-temperature data points of the electric drive system after the new energy vehicle starts, and the voltage-temperature data points include the motor voltage and motor temperature at each sampling moment;

[0007] According to the neighborhood distribution density and relative position of each voltage-temperature data point, obtain a preset number of reference clustering center points; according to the change in the data point density on the line connecting each voltage-temperature data point and each reference clustering center point, and the connection slope of each voltage-temperature data point relative to each reference clustering center point, obtain the clustering influence weight of each voltage-temperature data point and each reference clustering center point;

[0008] Perform clustering analysis according to the clustering influence weight and the Euclidean distance between each voltage-temperature data point and each reference clustering center point to obtain at least two voltage-temperature data clustering clusters; according to the data point density inside each voltage-temperature data clustering cluster and the relative outlier degree of the voltage-temperature data clustering cluster, obtain the clustering anomaly degree of each voltage-temperature data clustering cluster;

[0009] Monitor the driving state of the new energy vehicle electric drive system according to the clustering anomaly degree.

[0010] Further, the method for obtaining the reference clustering center point includes:

[0011] Take the normalized value of the Euclidean distance between each voltage-temperature data point and the voltage-temperature data point closest to it as the local sparsity degree of each voltage-temperature data; take the voltage-temperature data point with the smallest local sparsity degree as the reference clustering center point during the first iteration traversal;

[0012] In each subsequent iteration traversal after the first iteration traversal, according to the local sparsity degree of all voltage-temperature data points outside the reference clustering center point, and the clustering distance between each voltage-temperature data point and the reference clustering center point, obtain the clustering center possibility degree of each voltage-temperature data point during each iteration traversal;

[0013] For each iteration traversal, the voltage-temperature data point corresponding to the maximum degree of the cluster center among all voltage-temperature data points is used as the reference cluster center point for each iteration traversal; continue the iterative traversal until the number of reference cluster center points reaches the preset number of cluster clusters and then stop.

[0014] Furthermore, the method for obtaining the clustering influence weight includes:

[0015] In turn, each reference cluster center point is used as the target cluster center point; all voltage-temperature data points outside the reference cluster center point are used as analysis data points; all voltage-temperature data points on the line segment with each analysis data point and the target cluster center point as endpoints are used as the reference data points corresponding to each analysis data point;

[0016] Along the direction from each analysis data point to the target cluster center point, the Euclidean distance between each reference data point and the next reference data point is used as the reference interval distance of each reference data point; the negative correlation mapping value of the ratio of the variance of the reference interval distances of all reference data points to the number of reference data points is used as the reference density stability degree between each analysis data point and the target cluster center point;

[0017] The arctangent function value of the slope of the straight line between each analysis data point and the target cluster center point is used as the reference angle of each analysis data point; other analysis data points of each analysis data point are used as the comparison data points of each analysis data point; the difference between the reference angle of each comparison data point and the reference angle of the corresponding analysis data point is used as the comparison angle of each comparison data point; the negative correlation mapping value of the accumulated value of the comparison angles of all comparison data points is used as the overall angle consistency between each analysis data point and the target cluster center point;

[0018] According to the reference density stability degree and the overall angle consistency, the clustering influence weight between each analysis data point and the target cluster center point is obtained; both the reference density stability degree and the overall angle consistency are negatively correlated with the clustering influence weight.

[0019] Furthermore, the method for obtaining the voltage-temperature data cluster includes:

[0020] The product of the Euclidean distance between each voltage-temperature data point and each reference cluster center point and the clustering influence weight is used as the clustering distance between each voltage-temperature data point and each reference cluster center point; clustering analysis is performed according to the clustering distance through the k-means clustering algorithm to obtain at least two voltage-temperature data clusters.

[0021] Furthermore, the calculation formula for the clustering abnormality degree includes:

[0022]

[0023] Among them, E h is the clustering anomaly degree of the h-th voltage-temperature data clustering cluster, and N h is the number of voltage-temperature data points in the h-th voltage-temperature data clustering cluster; S h is the area of the minimum bounding rectangle of the h-th voltage-temperature data clustering cluster; M h is the number of other voltage-temperature data clustering clusters outside the h-th voltage-temperature data clustering cluster; d m,i is the Euclidean distance between the clustering center of the h-th voltage-temperature data clustering cluster and the clustering center of the i-th voltage-temperature data clustering cluster outside it; exp() is the exponential function with the natural constant as the base; tanh() is the hyperbolic tangent function.

[0024] Furthermore, the method for monitoring the driving state of the new energy vehicle electric drive system according to the clustering anomaly degree includes:

[0025] When there is a voltage-temperature data clustering cluster with a clustering anomaly degree greater than the preset anomaly threshold, the driving state of the new energy vehicle electric drive system is abnormal.

[0026] Furthermore, the method for obtaining the possibility degree of the clustering center includes:

[0027] For any iteration traversal except the first iteration traversal:

[0028] The normalized value of the accumulated Euclidean distance between each voltage-temperature data point and all reference clustering center points is used as the degree of separation between clusters of each voltage-temperature data point during the iteration traversal; the sum value of the local sparsity degree and the degree of separation between clusters is used as the possibility degree of the clustering center of each voltage-temperature data point during the iteration traversal.

[0029] Furthermore, the method for obtaining the clustering influence weight between each analysis data point and the target clustering center point according to the reference density stability degree and the overall angle consistency includes:

[0030] The negative correlation mapping value of the mean between the reference density stability degree and the overall angle consistency is used as the clustering influence weight between each analysis data point and the target clustering center point.

[0031] Furthermore, the preset anomaly threshold is set to 0.6.

[0032] Furthermore, the preset number of clustering clusters is set to 3.

[0033] The present invention has the following beneficial effects:

[0034] Since the voltage and temperature corresponding to the electric drive system after the new energy vehicle starts will show a slow increase in a linear trend, and the corresponding voltage-temperature data points will show a linearly positively correlated extended distribution trend. When performing clustering analysis only based on the Euclidean distance, abnormal voltage-temperature data points will be classified into the clustering clusters corresponding to normal data points, resulting in abnormal voltage-temperature data points being classified into the clustering clusters corresponding to normal data points, causing the clustering process to fall into a local optimum, thus affecting the accuracy of the clustering result. Therefore, in order to obtain a correct clustering result, it is necessary to classify abnormal voltage-temperature data points into the clustering clusters corresponding to abnormal data points as much as possible.

[0035] Among all the voltage-temperature data points of the electric drive system after the new energy vehicle starts, although the abnormal voltage-temperature data points are relatively close to the normal voltage-temperature data points in terms of Euclidean distance, the distribution trends of abnormal voltage-temperature data points and normal voltage-temperature data points are significantly different. Visually, the distribution of normal voltage-temperature data points is relatively dense and shows a certain positively correlated extended trend. In the case of abnormalities, the voltage-temperature data of the electric drive system will show two motor states of high temperature and low voltage and low temperature and high voltage, and are more discrete than normal data points. That is, when the clustering result is reasonable, there are three types of the largest clustering clusters, namely, the set of voltage-temperature data points with low temperature and high voltage, the set of voltage-temperature data points with high temperature and low voltage, and the set of normal voltage-temperature data points. Therefore, if the accuracy of the clustering result is to be improved, it is necessary to classify the voltage-temperature data points into the correct data point sets as much as possible.

[0036] The premise for dividing clustering clusters is to determine the clustering centers. Since the abnormal voltage-temperature data points are relatively outlying in the overall distribution and the neighboring data points corresponding to the abnormal voltage-temperature data points are relatively sparse, the neighboring data points of the data points at the clustering centers need to be as discrete as possible and the intervals between different clustering centers need to be as far as possible to avoid falling into a local optimum while ensuring that abnormal voltage-temperature data points with the same abnormal type are classified into separate clustering centers. Therefore, the present invention first obtains a preset number of reference clustering centers according to the neighborhood distribution density and relative positions of each voltage-temperature data point, so as to obtain the determined clustering centers.

[0037] Furthermore, for normal voltage-temperature data points, when a new energy vehicle starts and speeds up, as the vehicle runs, the corresponding voltage gradually increases, and the motor temperature also gradually rises; when the voltage decreases, the motor temperature also gradually decreases; that is, there is a strong correlation between normal voltage-temperature data points, which means that the distribution of normal voltage-temperature data points is relatively dense and shows a certain positive correlation extension trend. Therefore, the connections between each corresponding normal voltage-temperature data point and the clustering center of normal voltage-temperature data points should generally maintain a similar direction, and the distribution of voltage-temperature data points on the connection line is relatively dense and the distance intervals are relatively stable; based on this characteristic, by introducing weights to affect the clustering distance between normal voltage-temperature data points and the clustering center of normal voltage-temperature data points, the normal voltage-temperature data points can be divided into the same clustering cluster as much as possible;

[0038] The same applies to abnormal voltage-temperature data points. Since abnormal voltage-temperature data points do not conform to the voltage-temperature relationship of the motor, compared with normal voltage-temperature data points, the direction corresponding to the connection line between abnormal voltage-temperature data points and the clustering center of normal voltage-temperature data points is inconsistent with the connection line direction of most normal voltage data points; however, when the clustering center pointed to by all voltage-temperature data points is the clustering center of abnormal voltage-temperature data points, since the abnormal clustering center is located at a position where the data points are relatively discrete, the angular difference between the abnormal voltage-temperature data points and the normal voltage-temperature data points pointing to the clustering center will be relatively small; therefore, based on this characteristic, by introducing weights to affect the clustering distance between abnormal voltage-temperature data points and the clustering center of abnormal voltage-temperature data points, the abnormal voltage-temperature data points can be divided into the same clustering cluster as much as possible; therefore, according to the present invention, based on the change in the data point density on the connection line between each voltage-temperature data point and each reference clustering center point, and the connection line slope of each voltage-temperature data point relative to each reference clustering center point, a more accurate clustering influence weight of each voltage-temperature data point and each reference clustering center point is obtained; thereby making the clustering distance between the voltage-temperature data points and each reference clustering center point affected by the clustering influence weight, so that the voltage-temperature data clustering clusters obtained by clustering analysis are more reasonable;

[0039] Furthermore, according to the characteristics that abnormal voltage-temperature data points are relatively discrete in distribution and relatively far from other clustering clusters, based on the data point density within each voltage-temperature data clustering cluster and the relative outlier degree of the voltage-temperature data clustering cluster, a more accurate clustering abnormality degree of each voltage-temperature data clustering cluster is obtained; making the subsequent monitoring effect of the driving state of the electric drive system of the new energy vehicle based on the clustering abnormality degree better. Description of the Drawings

[0040] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0041] Figure 1 Flowchart of a method for monitoring the driving state of an electric drive system of a new energy vehicle based on machine learning provided by an embodiment of the present invention;

[0042] Figure 2 An image for analyzing the angle of voltage and temperature data points provided by an embodiment of the present invention. Detailed implementation manners

[0043] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation manners, structures, features and effects of a method for monitoring the driving state of an electric drive system of a new energy vehicle based on machine learning proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0045] The following specifically describes the specific solution of a method for monitoring the driving state of an electric drive system of a new energy vehicle based on machine learning provided by the present invention in conjunction with the accompanying drawings.

[0046] Please refer to Figure 1 , which shows a flowchart of a method for monitoring the driving state of an electric drive system of a new energy vehicle based on machine learning provided by an embodiment of the present invention. The method includes:

[0047] Step S1: Obtain all voltage and temperature data points of the electric drive system after the new energy vehicle starts. The voltage and temperature data points include the motor voltage and motor temperature at each sampling moment.

[0048] An embodiment of the present invention aims to provide a method for monitoring the driving state of an electric drive system of a new energy vehicle based on machine learning, which is used to analyze the motor voltage and motor temperature of the electric drive system after the new energy vehicle is started. By improving the clustering distance in the clustering analysis process, a more accurate clustering analysis result can be obtained; thus, the monitoring effect of the driving state of the electric drive system of the new energy vehicle according to the clustering abnormality degree of each clustering cluster in the clustering analysis result is better; therefore, it is first necessary to obtain voltage data and temperature data.

[0049] Considering that the embodiment of the present invention needs to be analyzed by the method of clustering analysis, in order to make the analysis process more intuitive, the embodiment of the present invention obtains all voltage-temperature data points of the electric drive system after the new energy vehicle is started. The voltage-temperature data points include the motor voltage and motor temperature at each sampling moment. In the embodiment of the present invention, the motor temperature data is collected by setting a temperature sensor on the surface of the motor, and the motor voltage data is collected by setting a voltage sensor on the surface of the motor. And the motor temperature data and the motor voltage data are collected simultaneously at each sampling moment; in the embodiment of the present invention, the motor voltage is used as the horizontal axis and the motor temperature is used as the vertical axis, so as to intuitively obtain the voltage-temperature data point corresponding to each sampling moment. In the embodiment of the present invention, 100 voltage-temperature data points at sampling moments are collected for analysis after the new energy vehicle is started, and the time interval between sampling moments is set to 1 minute. The implementer can adjust the number of voltage-temperature data points and the time interval between sampling moments according to the specific situation, and no further elaboration is made here.

[0050] Step S2: Obtain the number of reference clustering center points of the preset clustering clusters according to the density of the neighborhood distribution and the relative position of each voltage-temperature data point; obtain the clustering influence weight of each voltage-temperature data point and each reference clustering center point according to the change of the data point density on the line connecting each voltage-temperature data point and each reference clustering center point, and the connection slope of each voltage-temperature data point relative to each reference clustering center point.

[0051] Since the voltage and temperature corresponding to the electric drive system after the new energy vehicle is started will show a linear trend of slow increase, and the corresponding voltage-temperature data points will show a linearly positively correlated extended distribution trend, when clustering analysis is performed only based on the Euclidean distance, the abnormal voltage-temperature data points will be divided into the clustering clusters corresponding to the normal data points, resulting in the abnormal voltage-temperature data points being divided into the clustering clusters corresponding to the normal data points, making the clustering process fall into a local optimal solution, thus affecting the accuracy of the clustering result; therefore, in order to obtain a correct clustering result, it is necessary to divide the abnormal voltage-temperature data points into the clustering clusters corresponding to the abnormal data points as much as possible.

[0052] Among all the voltage and temperature data points of the electric drive system after the new energy vehicle is started, although the abnormal voltage and temperature data points are relatively close to the normal voltage and temperature data points in terms of Euclidean distance, the distribution trends of the abnormal voltage and temperature data points and the normal voltage and temperature data points are significantly different. Visually, the distribution of the normal voltage and temperature data points is relatively dense and shows a certain positive correlation extension trend; in the case of abnormality, the voltage and temperature data of the electric drive system will show two motor states of high temperature and low voltage and low temperature and high voltage, and are more discrete than the normal data points; that is, when the clustering result is reasonable, there are three types of the largest clustering clusters, namely, the set of voltage and temperature data points with low temperature and high voltage, the set of voltage and temperature data points with high temperature and low voltage, and the set of normal voltage and temperature data points; therefore, if the accuracy of the clustering result is to be improved, it is necessary to divide the voltage and temperature data points into the correct data point sets as much as possible, that is, the purpose of the embodiments of the present invention.

[0053] The premise for dividing the clustering clusters is to determine the clustering centers. Since the abnormal voltage and temperature data points are relatively outlier in the overall distribution and the distribution of the neighboring data points corresponding to the abnormal voltage and temperature data points is relatively sparse, the neighboring data points of the data points at the clustering centers need to be as discrete as possible and the intervals between different clustering centers need to be as far as possible, so as to avoid falling into the local optimal solution and ensure that the abnormal voltage and temperature data points with the same abnormal type are divided into separate clustering centers; therefore, the embodiments of the present invention obtain a preset number of reference clustering centers according to the neighborhood distribution density and relative positions of each voltage and temperature data point. Preferably, the preset number of clustering clusters is set to 3. Since there are three types of the largest clustering clusters when the clustering result is reasonable, the preset number of clustering clusters is set to 3. The implementer can adjust the size of the preset number of clustering clusters according to the specific implementation environment, and no further elaboration will be made here.

[0054] Preferably, the method for obtaining the reference clustering centers includes:

[0055] The normalized value of the Euclidean distance between each voltage and temperature data point and the voltage and temperature data point closest to it is used as the local sparsity degree of each voltage and temperature data; the voltage and temperature data point with the smallest local sparsity degree is used as the reference clustering center during the first iteration traversal. Since the neighboring data points of the data points at the clustering centers need to be as discrete as possible and the intervals between different clustering centers need to be as far as possible, it is necessary to measure the possibility of the reference clustering centers based on the density of the neighboring data points of each voltage and temperature data point and the distance between each voltage and temperature data point and each clustering center; and the larger the density of the local data points of a voltage and temperature data point, the smaller the Euclidean distance between it and the voltage and temperature data point closest to it. Therefore, the voltage and temperature data point with the smallest local sparsity degree is selected as the reference clustering center during the first iteration traversal.

[0056] However, in the initial situation, there is no clustering center point. Therefore, in the embodiments of the present invention, only the density of the neighborhood data points of each voltage-temperature data point can be used, that is, the first reference clustering center point is obtained by the distance between the voltage-temperature data point and the voltage-temperature data point closest to it. Since the possibility of the reference clustering center point of each voltage-temperature data point needs to be measured, the density of the neighborhood data points of each voltage-temperature data point and the distance between each voltage-temperature data point and each clustering center point need to be considered. Therefore, after each reference clustering center point is obtained, the distance between the voltage-temperature data point and each clustering center point will change. Therefore, it is necessary to re-iterate to calculate the possible degree of the reference clustering center point. Therefore, in the first iteration traversal and each subsequent iteration traversal in the embodiments of the present invention, according to the local sparsity degree of all voltage-temperature data points outside the reference clustering center point and the clustering distance between each voltage-temperature data point and the reference clustering center point, the possible degree of the clustering center of each voltage-temperature data point is obtained for each iteration traversal.

[0057] Preferably, the method for obtaining the possible degree of the clustering center includes:

[0058] For any iteration traversal except the first iteration traversal:

[0059] The normalized value of the accumulated Euclidean distance between each voltage-temperature data point and all reference clustering center points is used as the degree of separation between clusters of each voltage-temperature data point during the iteration traversal; the sum of the local sparsity degree and the degree of separation between clusters is used as the possible degree of the clustering center of each voltage-temperature data point during the iteration traversal. Since the possibility of the reference clustering center point of each voltage-temperature data point needs to be measured, the density of the neighborhood data points of each voltage-temperature data point and the distance between each voltage-temperature data point and each clustering center point need to be considered; and the local sparsity degree can represent the density of the neighborhood data points of each voltage data point. Therefore, further, the degree of separation between clusters is used to represent the distance between each voltage-temperature data point and each clustering center point, and the possible degree of the clustering center is obtained through the sum of the local sparsity degree and the degree of separation between clusters.

[0060] Since only one reference clustering center point is selected in each iteration traversal, in the embodiments of the present invention, the voltage-temperature data point corresponding to the maximum clustering center possibility among all voltage-temperature data points during each iteration traversal is used as the reference clustering center point during each iteration traversal; continue the iterative traversal until the number of reference clustering center points reaches the preset number of clustering clusters and then stop. Since the preset number of clustering cluster data in the embodiments of the present invention is set to 3, after obtaining two reference clustering center points, continue to recalculate the inter-cluster separation degree of each voltage-temperature data point, and obtain the clustering center possibility during the third iteration traversal according to the sum of the inter-cluster separation degree and the local sparsity degree of each voltage-temperature data point, and use the reference clustering center point corresponding to the maximum clustering center possibility during the third iteration traversal as the reference clustering center point during the third iteration traversal. It should be noted that during each iteration traversal of the present invention, the calculation of the clustering center possibility is not performed on all previous reference clustering center points, and no further elaboration will be made here.

[0061] In the embodiments of the present invention, each voltage-temperature data point during each iteration traversal subsequent to the first outer iteration traversal is sequentially used as the k-th voltage-temperature data point during the r-th iteration traversal; then the method for obtaining the clustering center possibility of the k-th voltage-temperature data point during the r-th iteration traversal is expressed in formula as:

[0062]

[0063] Where, D r,k is the clustering center possibility of the k-th voltage-temperature data point during the r-th iteration traversal; d k,min is the Euclidean distance between the k-th voltage-temperature data point and the voltage-temperature data point closest to it, Norm(d k,min ) is the local sparsity degree of the k-th voltage-temperature data point; d k,j is the Euclidean distance between the k-th voltage-temperature data point and the j-th reference clustering center point during the r-th iteration traversal; r is the number of iteration traversals, so r - 1 is the number of reference clustering center points during the r-th iteration traversal; is the inter-cluster interval degree of the k-th voltage-temperature data point during the r-th iteration traversal; Norm() is the normalization function, and the normalization method of the normalization function in the embodiments of the present invention all selects linear normalization, and the implementer can select other normalization methods according to the specific implementation environment, and no further elaboration will be made here.

[0064] Further, for normal voltage-temperature data points, when a new energy vehicle starts and speeds up, as the vehicle runs, the corresponding voltage gradually increases, and the motor temperature also gradually rises; when the voltage decreases, the motor temperature also gradually decreases; that is, there is a strong correlation between normal voltage-temperature data points, which means that the distribution of normal voltage-temperature data points is relatively dense and shows a certain positive correlation extension trend. Therefore, the connection lines between each corresponding normal voltage-temperature data point and the clustering center of normal voltage-temperature data points should basically maintain a similar direction, and the distribution of voltage-temperature data points on the connection line is relatively dense and the distance interval is relatively stable; according to this characteristic, by introducing weights to affect the clustering distance between normal voltage-temperature data points and the clustering center of normal voltage-temperature data points, the normal voltage-temperature data points can be divided into the same clustering cluster as much as possible;

[0065] The same is true for abnormal voltage-temperature data points. Since abnormal voltage-temperature data points are points that do not conform to the voltage-temperature relationship of the motor, compared with normal voltage-temperature data points, the direction corresponding to the connection line between abnormal voltage-temperature data points and the clustering center of normal voltage-temperature data points is inconsistent with the connection line direction of most normal voltage data points; however, when the clustering center pointed to by all voltage-temperature data points is the clustering center of abnormal voltage-temperature data points, since the abnormal clustering center is located at a position where the data points are relatively discrete, the angular difference between the abnormal voltage-temperature data points and the normal voltage-temperature data points pointing to the clustering center will be relatively small; therefore, according to this characteristic, by introducing weights to affect the clustering distance between abnormal voltage-temperature data points and the clustering center of abnormal voltage-temperature data points, the abnormal voltage-temperature data points can be divided into the same clustering cluster as much as possible; therefore, in the embodiment of the present invention, according to the change in the data point density on the connection line between each voltage-temperature data point and each reference clustering center point, and the connection line slope of each voltage-temperature data point relative to each reference clustering center point, the clustering influence weight of each voltage-temperature data point and each reference clustering center point is obtained.

[0066] Preferably, the method for obtaining the clustering influence weight includes:

[0067] Successively take each reference clustering center point as the target clustering center point; take all voltage-temperature data points outside the reference clustering center point as analysis data points; take all voltage-temperature data points on the line segment with each analysis data point and the target clustering center point as endpoints as the reference data points corresponding to each analysis data point; along the direction from each analysis data point to the target clustering center point, take the Euclidean distance between each reference data point and the next reference data point as the reference interval distance of each reference data point; take the negative correlation mapping value of the ratio of the variance of the reference interval distances of all reference data points to the number of reference data points as the reference density stability degree between each analysis data point and the target clustering center point. Since the voltage-temperature data points are more densely distributed and the distance intervals are more stable on the line connecting each normal voltage-temperature data point and the clustering center of normal voltage-temperature data points; therefore, by introducing the variance of the reference interval distances of all reference data points, the stability degree of the distance interval distribution on the corresponding line can be characterized; similarly, the more the number of reference data points, the denser the distribution of the corresponding voltage-temperature data points, that is, the number of reference data points can characterize the corresponding density distribution. And the smaller the variance and the more the number of pixel points, the stronger the corresponding correlation. Therefore, take the negative correlation mapping value of the ratio of the variance of the reference interval distances of all reference data points to the number of reference data points as the reference density stability degree between each analysis data point and the target clustering center point.

[0068] It should be noted that the purpose of distinguishing between comparison data points and analysis data points is to distinguish the analysis objects of data points. Since the reference clustering center points do not participate in the division of clustering clusters, all voltage-temperature data points outside the reference clustering center points are taken as analysis data points; and the analysis data points need to be compared and analyzed with other analysis data points. Therefore, take the other analysis data points of each analysis data point as the comparison data points of each analysis data point, making the subsequent analysis clearer after the distinction.

[0069] Take the arctangent function value of the slope of the straight line between each analysis data point and the target clustering center point as the reference angle of each analysis data point; take the other analysis data points of each analysis data point as the comparison data points of each analysis data point; take the difference between the reference angle of each comparison data point and the reference angle of the corresponding analysis data point as the comparison angle of each comparison data point; take the negative correlation mapping value of the cumulative value of the comparison angles of all comparison data points as the overall angle consistency between each analysis data point and the target clustering center point.

[0070] Please refer to Figure 2 which shows an angle analysis image of voltage-temperature data points provided by an embodiment of the present invention; in Figure 2Among them, the voltage-temperature data points with relatively dense distribution and showing a certain positive correlation extension trend are normal voltage-temperature data points. Among them, b, c, and d are all normal voltage-temperature data points, while a is an abnormal voltage-temperature data point with relatively discrete distribution; A is the clustering center of abnormal voltage-temperature data points, and B is the clustering center of normal voltage-temperature data points.

[0071] For normal voltage-temperature data points, the distribution of voltage-temperature data points is relatively dense and shows a certain positive correlation extension trend. Therefore, the connecting lines between each corresponding normal voltage-temperature data point and the clustering center of normal voltage-temperature data points should basically maintain a similar direction. In terms of slope, it shows that the linear slopes between normal voltage-temperature data points and the corresponding clustering center points are basically the same, which is reflected in Figure 2 that the slopes corresponding to the connecting lines of b, c, and d to B are basically the same respectively; since normal voltage-temperature data points account for the majority, the corresponding comparison angles are usually relatively small, that is, there are only relatively large reference angle differences between normal voltage-temperature data points and some discrete abnormal voltage-temperature data points; however, relatively speaking, when the target clustering center point pointed to by normal voltage-temperature data points is not the clustering center of normal voltage-temperature data points, the corresponding reference angle difference is relatively large. For example, in Figure 2 it, the slope differences corresponding to the connecting lines of b, c, and d to A are relatively large; therefore, normal voltage-temperature data points can be divided into the correct clustering clusters according to the comparison angles of each voltage-temperature data point; when normal voltage-temperature data points point to the clustering center of normal voltage-temperature data points, the corresponding overall comparison angle is smaller.

[0072] Although abnormal voltage-temperature data points do not have the characteristics of positive correlation extension and dense distribution, the comparison angle corresponding to when abnormal voltage-temperature data points point to the clustering center point of normal voltage-temperature data points is larger than that when pointing to the clustering center point of abnormal voltage data points. In Figure 2 it, it is known that the slope differences corresponding to the connecting lines of b, c, and d to A are relatively large; while the slopes corresponding to the connecting lines of b, c, and d to B are basically the same; and the slope differences between a and both A and B are relatively large; then relatively speaking, since normal voltage-temperature data points account for the majority, the overall comparison angle corresponding to when a points to A is relatively larger; and because the slope distributions corresponding to the connecting lines of b, c, and d to A are relatively chaotic, a has a similar reference angle to some normal abnormal voltage-temperature data points, that is, the comparison angle corresponding to when abnormal voltage-temperature data points point to the clustering center of abnormal voltage-temperature data points is relatively smaller overall. That is, the smaller the cumulative value of the comparison angle corresponding to each voltage-temperature data point and the target clustering center point, the more likely it belongs to a clustering cluster. That is, the higher the overall angle consistency, so the cumulative value of the comparison angle is negatively correlated.

[0073] According to the reference density stability degree and the overall angle consistency, the clustering influence weight between each analysis data point and the target clustering center point is obtained; both the reference density stability degree and the overall angle consistency have a negative correlation with the clustering influence weight. Since when the reference density stability degree is larger, normal voltage-temperature data points are more likely to be divided into a clustering cluster; while for any clustering center, the corresponding reference density stability degree of abnormal voltage-temperature data points is smaller; therefore, the reference density stability degree can, to a certain extent, characterize the belonging relationship of the clustering cluster of normal voltage-temperature data points and does not affect the clustering cluster division of abnormal voltage-temperature data points. And when the overall angle consistency is larger, the corresponding voltage-temperature data points are more likely to belong to the clustering cluster corresponding to the target clustering center point. Therefore, the reference density stability degree and the overall angle consistency can be combined to jointly represent the clustering influence weight. Since the basis of clustering analysis is the distance between voltage-temperature data points and the clustering center, if we want the voltage-temperature data points to be clustered into the target clustering center point as much as possible, then we need to reduce the clustering influence weight for one to reduce the distance required for clustering analysis; therefore, both the reference density stability degree and the overall angle consistency have a negative correlation with the clustering influence weight.

[0074] Preferably, the method for obtaining the clustering influence weight between each analysis data point and the target clustering center point according to the reference density stability degree and the overall angle consistency includes:

[0075] Taking the negative correlation mapping value of the mean between the reference density stability degree and the overall angle consistency as the clustering influence weight between each analysis data point and the target clustering center point. It should be noted that the implementer can also obtain the clustering influence weight according to the reference density stability degree and the overall angle consistency through other methods. For example, taking the normalized value of the product between the negative correlation mapping value of the reference density stability degree and the negative correlation mapping value of the overall angle consistency as the clustering influence weight, which will not be elaborated further here.

[0076] In the embodiment of the present invention, each analysis data point is sequentially taken as the u-th analysis data point, and each reference clustering center point is sequentially taken as the v-th reference clustering center point; then the method for obtaining the clustering influence weight between the u-th analysis data point and the v-th reference clustering center point is expressed in the formula as:

[0077]

[0078] where, R u,v is the clustering influence weight between the u-th analysis data point and the v-th reference clustering center point; s u,v is the reference interval distance variance of all reference data points in the direction from the u-th analysis data point to the v-th reference clustering center point; n u,vis the number of reference data points in the direction from the u-th analysis data point to the v-th reference cluster center point; is the reference density stability between the u-th analysis data point and the v-th reference cluster center point; f' ut,v is the slope of the straight line between the t-th comparison data point corresponding to the u-th analysis data point and the v-th reference cluster center point, arctan(f' ut,v ) is the reference angle of the t-th comparison data point corresponding to the u-th analysis data point under the v-th reference cluster center point; f' u,v The slope of the straight line between the u-th analysis data point and the v-th reference cluster center point, arctan(f' u,v ) is the reference angle of the u-th analysis data point under the v-th reference cluster center point; |arctan(f' ut,v ) - arctan(f' u,v )| is the comparison angle of the t-th comparison data point corresponding to the u-th analysis data point under the v-th reference cluster center point; is the overall angle consistency between the u-th analysis data point and the v-th reference cluster center point; exp() is the exponential function with the natural constant as the base, arctan() is the arctangent function, and || is the absolute value symbol. It should be noted that in addition to performing negative correlation mapping through a 1 - minus function, the implementer can also perform negative correlation mapping through other methods, such as the exp(-) function.

[0079] Step S3: According to the clustering influence weight and the Euclidean distance between each voltage - temperature data point and each reference cluster center point, perform clustering analysis to obtain at least two voltage - temperature data clustering clusters; according to the data point density within each voltage - temperature data clustering cluster and the relative outlier degree of the voltage - temperature data clustering cluster, obtain the clustering anomaly degree of each voltage - temperature data clustering cluster.

[0080] After obtaining the clustering influence weight, further according to the purpose of the embodiments of the present invention, adjust the clustering distance between each voltage - temperature data point and each reference cluster center point according to the clustering influence weight. The embodiments of the present invention perform clustering analysis according to the clustering influence weight and the Euclidean distance between each voltage - temperature data point and each reference cluster center point to obtain at least two voltage - temperature data clustering clusters.

[0081] Preferably, the method for obtaining the voltage - temperature data clustering cluster includes:

[0082] When the clustering influence weight is smaller, the corresponding voltage-temperature data points are more likely to belong to the clustering cluster of the reference clustering center point. Therefore, in the embodiments of the present invention, the product of the Euclidean distance between each voltage-temperature data point and each reference clustering center point and the clustering influence weight is used as the clustering distance between each voltage-temperature data point and each reference clustering center point; by reducing the clustering influence weight, the clustering distance between the voltage-temperature data points with a higher degree of belonging and the reference clustering center is reduced, making the obtained clustering analysis result more accurate after division. Further, clustering analysis is performed according to the clustering distance by using the k-means clustering algorithm to obtain at least two voltage-temperature data clustering clusters. It should be noted that the k-means clustering algorithm is a well-known prior art to those skilled in the art and will not be further described herein. Implementers can also adopt other clustering methods according to the specific implementation environment.

[0083] Further, when the clustering result is reasonable, there are three types of the largest clustering cluster types, that is, the set of voltage-temperature data points with low temperature and high voltage, the set of voltage-temperature data points with high temperature and low voltage, and the set of normal voltage-temperature data points; while the sets of abnormal voltage-temperature data points with low temperature and high voltage and high temperature and low voltage are usually distributed on both sides of the set of normal voltage-temperature data points. Therefore, the clustering clusters of abnormal voltage-temperature data points are relatively far from other clustering clusters; and the number of abnormal voltage-temperature data points is usually small, so the data point density in the clustering clusters of abnormal voltage-temperature data points is small; therefore, in the embodiments of the present invention, the clustering abnormality degree of each voltage-temperature data clustering cluster is obtained according to the data point density inside each voltage-temperature data clustering cluster and the relative outlier degree of the voltage-temperature data clustering cluster;

[0084] Preferably, each voltage-temperature data clustering cluster is sequentially used as the h-th voltage-temperature data clustering cluster, and the calculation formula for the clustering abnormality degree of the h-th voltage-temperature data clustering cluster includes:

[0085]

[0086] wherein, E h is the clustering abnormality degree of the h-th voltage-temperature data clustering cluster, N h is the number of voltage-temperature data points in the h-th voltage-temperature data clustering cluster; S h is the area of the minimum circumscribed rectangle of the h-th voltage-temperature data clustering cluster; M h is the number of other voltage-temperature data clustering clusters outside the h-th voltage-temperature data clustering cluster; d m,i is the Euclidean distance between the clustering center of the h-th voltage-temperature data clustering cluster and the clustering center of the i-th voltage-temperature data clustering cluster outside it; exp() is the exponential function with the natural constant as the base; tanh() is the hyperbolic tangent function.

[0087] In the calculation formula of the clustering anomaly degree, by and The mean value of represents the clustering anomaly degree. First, for each voltage-temperature data clustering cluster, the clustering cluster of abnormal voltage-temperature data points is farther away from other clustering clusters, and the density of data points in the clustering cluster of abnormal voltage-temperature data points is smaller; therefore, the smaller the density of data points in the voltage-temperature data clustering cluster and the farther the distance from other clustering clusters, the greater the anomaly degree of the corresponding clustering cluster. And is the ratio of the number of voltage-temperature data points in the voltage-temperature data clustering cluster to the area of the corresponding minimum bounding rectangle. The larger the corresponding ratio, the denser the data points in the corresponding voltage-temperature data clustering cluster, that is, the smaller the clustering anomaly degree; therefore, through exp(-) for Perform a negative correlation mapping. Implementers can also select other negative correlation mapping methods according to the specific implementation environment, such as the reciprocal, etc., which will not be elaborated further here. Since the distance between the voltage-temperature data clustering cluster and other clustering clusters can also represent the clustering anomaly degree, the present invention accumulates the distances between the clustering centers corresponding to each voltage-temperature data clustering cluster and the clustering centers of each other clustering cluster, and performs positive correlation normalization through the tanh() function to make the corresponding values more reasonable. Implementers can also adopt other positive correlation normalization methods according to the specific implementation environment, which will not be elaborated further here.

[0088] Step S4: Monitor the driving state of the new energy vehicle electric drive system according to the clustering anomaly degree.

[0089] After obtaining the clustering anomaly degree of each voltage-temperature data clustering cluster, further, the driving state of the new energy vehicle electric drive system can be monitored according to the clustering anomaly degree.

[0090] Preferably, the method for monitoring the driving state of the new energy vehicle electric drive system according to the clustering anomaly degree includes:

[0091] Since the greater the clustering anomaly degree, the more likely the corresponding voltage-temperature data clustering cluster is the clustering cluster corresponding to the abnormal voltage-temperature data points, and the existence of the abnormal voltage-temperature data points can indicate that the driving state of the new energy vehicle electric drive system is abnormal. Therefore, when there is a voltage-temperature data clustering cluster with a large clustering anomaly degree, it indicates that the driving state of the new energy vehicle electric drive system is abnormal. In the embodiment of the present invention, when there is a voltage-temperature data clustering cluster with a clustering anomaly degree greater than the preset anomaly threshold, the driving state of the new energy vehicle electric drive system is abnormal. Preferably, the preset anomaly threshold is set to 0.6. Implementers can adjust the size of the preset anomaly threshold according to the specific implementation environment, which will not be elaborated further here.

[0092] In summary, the present invention first screens the reference clustering center points according to the density of voltage-temperature data points in the neighborhood distribution and their relative positions; further, according to the change in the density of data points on the line connecting the voltage-temperature data points and the reference clustering center points, and the slope of the connection line, the clustering influence weight is obtained; the clustering influence weight affects the clustering distance during clustering, so that abnormal voltage-temperature data points can be more likely to be assigned to abnormal voltage-temperature data clustering clusters, that is, a more accurate clustering analysis result is obtained; thereby, a more accurate clustering abnormality degree can be obtained according to the density of data points and the relative outlier situation of the clustering clusters; and the effect of monitoring the driving state of the electric drive system of new energy vehicles according to the clustering abnormality degree is better.

[0093] It should be noted that the above sequence of embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0094] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and the key points of each embodiment are the differences from other embodiments.

Claims

1. A driving state monitoring method for electric drive system of new energy vehicle based on machine learning, characterized in that: The method comprises: Acquire all voltage and temperature data points of the electric drive system after the new energy vehicle is started, wherein the voltage and temperature data points include the motor voltage and the motor temperature at each sampling moment; According to the local sparsity degree of each voltage and temperature data point and the degree of separation between clusters, a preset number of reference cluster center points of cluster clusters are obtained; according to the density change of data points on the line between each voltage and temperature data point and each reference cluster center point, and the slope of the line between each voltage and temperature data point relative to each reference cluster center point, the clustering influence weight of each voltage and temperature data point and each reference cluster center point is obtained; the normalized value of the Euclidean distance between each voltage and temperature data point and the voltage and temperature data point closest to it is used as the local sparsity degree of each voltage and temperature data point; the normalized value of the accumulated value of the Euclidean distance between each voltage and temperature data point and all reference cluster center points is used as the degree of separation between clusters; Perform cluster analysis according to the clustering influence weight and the Euclidean distance between each voltage and temperature data point and each reference cluster center point to obtain at least two voltage and temperature data clusters; obtain the clustering abnormality degree of each voltage and temperature data cluster according to the density of data points within each voltage and temperature data cluster and the relative outlier degree of the voltage and temperature data cluster; The driving state of the electric drive system of the new energy vehicle is monitored according to the degree of clustering abnormality.

2. The method for monitoring driving state of electric drive system of new energy vehicle based on machine learning according to claim 1 is characterized in that: The method for obtaining the reference cluster center point includes: The voltage-temperature data point with the smallest local sparsity is used as the reference cluster center point in the first iteration; In each subsequent iteration after the first iteration, the possible degree of the cluster center of each voltage and temperature data point in each iteration is obtained according to the local sparsity of all voltage and temperature data points outside the reference cluster center point and the degree of separation between clusters of each voltage and temperature data point; The voltage-temperature data point corresponding to the maximum possible degree of cluster center among all voltage-temperature data points in each iterative traversal is used as the reference cluster center point in each iterative traversal; the iterative traversal is continued until the number of reference cluster center points reaches the preset number of cluster clusters.

3. The method for monitoring driving state of electric drive system of new energy vehicle based on machine learning according to claim 1 is characterized in that: The method for obtaining the clustering influence weight includes: Each reference cluster center point is used as the target cluster center point in turn; all voltage and temperature data points outside the reference cluster center point are used as analysis data points; all voltage and temperature data points on the line segment with each analysis data point and the target cluster center point as endpoints are used as reference data points corresponding to each analysis data point; Along the direction from each analysis data point to the target cluster center point, the Euclidean distance between each reference data point and the next reference data point is used as the reference interval distance of each reference data point; the negative correlation mapping value of the ratio of the variance of the reference interval distance of all reference data points to the number of reference data points is used as the reference density stability between each analysis data point and the target cluster center point; The inverse tangent function value of the slope of the straight line between each analysis data point and the target cluster center point is used as the reference angle of each analysis data point; the other analysis data points of each analysis data point are used as the comparison data points of each analysis data point; the difference between the reference angle of each comparison data point and the reference angle of the corresponding analysis data point is used as the comparison angle of each comparison data point; the negative correlation mapping value of the accumulated value of the comparison angle of all comparison data points is used as the overall angle consistency between each analysis data point and the target cluster center point; According to the reference density stability and the overall angle consistency, the clustering influence weight between each analysis data point and the target cluster center point is obtained; the reference density stability and the overall angle consistency are negatively correlated with the clustering influence weight.

4. The method for monitoring driving state of electric drive system of new energy vehicle based on machine learning according to claim 1, characterized in that: The method for obtaining the voltage and temperature data clusters includes: The Euclidean distance between each voltage-temperature data point and each reference cluster center point is multiplied by the cluster influence weight as the cluster distance between each voltage-temperature data point and each reference cluster center point; cluster analysis is performed using a k-means clustering algorithm based on the cluster distance to obtain at least two voltage-temperature data clusters.

5. The method for monitoring driving state of electric drive system of new energy vehicle based on machine learning according to claim 1, characterized in that: The calculation formula of the clustering abnormality degree includes: in, For the The clustering abnormality degree of the voltage and temperature data clusters, For the The number of voltage and temperature data points in a voltage and temperature data cluster; For the The area of ​​the minimum circumscribed rectangle of the voltage and temperature data clusters; For the The number of other voltage-temperature data clustering clusters outside the voltage-temperature data clustering cluster; For the The cluster center of the voltage and temperature data cluster is compared with the cluster center of the other The Euclidean distance between the cluster centers of the voltage and temperature data clusters; is an exponential function with a natural constant as base; is the hyperbolic tangent function.

6. The method for monitoring driving state of electric drive system of new energy vehicle based on machine learning according to claim 5 is characterized in that: The method for monitoring the driving state of the electric drive system of a new energy vehicle according to the clustering abnormality degree includes: When there is a voltage and temperature data cluster whose clustering abnormality is greater than a preset abnormality threshold, the driving state of the new energy vehicle electric drive system is abnormal.

7. The method for monitoring driving state of electric drive system of new energy vehicle based on machine learning according to claim 2 is characterized in that: The method for obtaining the possible degree of the cluster center includes: For any iteration except the first one: The sum of the local sparsity degree and the inter-cluster spacing degree is used as the possible degree of the cluster center of each voltage-temperature data point during iterative traversal.

8. The method for monitoring driving state of electric drive system of new energy vehicle based on machine learning according to claim 3 is characterized in that: The method for obtaining the clustering influence weight between each analysis data point and the target cluster center point according to the reference density stability and the overall angle consistency includes: The negative correlation mapping value of the mean between the reference density stability and the overall angle consistency is used as the clustering influence weight between each analysis data point and the target cluster center point.

9. The method for monitoring driving state of electric drive system of new energy vehicle based on machine learning according to claim 6 is characterized in that: The preset abnormal threshold is set to 0.

6.

10. The method for monitoring driving state of electric drive system of new energy vehicle based on machine learning according to claim 1, characterized in that: The preset number of clusters is set to 3.

Images