A driving behavior primitive clustering method considering data variation trend

Abstract

The application relates to a driving behavior primitive clustering method considering data change trends, comprising the following steps: step one: collecting natural driving data, selecting driving behavior analysis variables, obtaining a driving behavior data set through preprocessed driving behavior data and driving behavior derivative data after filtering processing; step two: extracting driving behavior primitives by using a BMASS method to obtain a driving behavior primitive set P={P j ;} j＝1:N ; step three: preliminarily discretizing the driving behavior primitive data; step four: judging the data change trends of the driving behavior primitives according to the derivatives and data fluctuation conditions of t points and the neighborhoods of the t points in P j ; step five: obtaining the discretization results of the driving behavior primitives by comprehensively considering data distribution items and data change trend items, and realizing the clustering of the driving behavior primitives based on an LDA topic model; the application can improve the clustering accuracy of the driving behavior primitives, realize the accurate expression of semantic information of the driving behavior primitives, and help drivers accurately and intuitively understand their driving behaviors.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of driving behavior primitive clustering, in particular to a driving behavior primitive clustering method considering data variation trend. BACKGROUND

[0002] Driving behavior primitive is the basic component unit of driving behavior, which is the smallest driving behavior data segment with physical meaning. Although multiple independent driving behavior primitives can be obtained in the primitive extraction link, the types of driving behavior primitives are limited. In order to intuitively understand the driving behavior primitive, it is necessary to cluster the driving behavior primitive.

[0003] LDA topic model is commonly used for text clustering. Using LDA model for primitive clustering first needs to discretize continuous driving behavior data. Since the bag-of-words model is used in LDA topic model, the order of word appearance is not considered, so when using the model to cluster driving behavior primitives, the time characteristics of driving behavior primitives will be ignored, resulting in unreasonable results, which makes it impossible to accurately cluster driving behavior primitives, and thus leads to incorrect judgment of driving behavior state;

[0004] Therefore, the present application provides a driving behavior primitive clustering method considering driving behavior data variation trend. After obtaining the data distribution item of driving behavior data based on semantic space, a data variation trend item is added to make up for the deficiency of LDA model in clustering time series data caused by bag-of-words model, so as to accurately cluster driving behavior primitives and accurately identify driving behavior state. SUMMARY

[0005] The present application provides a driving behavior primitive clustering method considering data variation trend, which includes the following steps: step one: collecting natural driving data of M drivers, selecting longitudinal driving behavior parameters speed v and longitudinal acceleration ax, and lateral driving behavior parameters lateral acceleration ay and yaw angle r as driving behavior analysis variables, using sliding average value to filter driving behavior analysis variable data to obtain preprocessed driving behavior data; deriving the preprocessed driving behavior data to obtain driving behavior derivative data;

[0006] The driving behavior data set of M drivers is obtained by preprocessed driving behavior data and driving behavior derivative data The The driving behavior data of the i-th driver is NM iData length of the i-th driver, the pre-processed driving behavior data includes speed v, longitudinal acceleration ax, lateral acceleration ay, and yaw angle r, and the driving behavior derivative data includes speed v derivative dv, longitudinal acceleration ax derivative dax, lateral acceleration ay derivative day, and yaw angle r derivative dr;

[0007] Step two: based on the pre-processed driving behavior data, the driving behavior data set is processed by using the BMASS method Driving behavior primitive extraction is performed to obtain a driving behavior primitive set P = {P j ;} j＝1:N N is the total number of primitives; The j-th driving behavior primitive, NN j Data length of the j-th driving behavior primitive;

[0008] Step three: constructing a semantic space, the driving behavior primitive data is preliminarily discretized to obtain a data distribution item output code of the driving behavior primitive;

[0009] Step four: based on the driving behavior primitive P j The data change trend of the driving behavior primitive is judged based on the derivative and data fluctuation of the t point and its neighborhood to obtain a data change trend item output code of the driving behavior primitive;

[0010] The data at the t point includes four driving behavior variables of speed v, longitudinal acceleration ax, lateral acceleration ay, and yaw angle r;

[0011] Step five: the data distribution item output code and the data change trend item output code of the driving behavior primitive are integrated to obtain a discretization result of the driving behavior primitive, the discretization result includes the distribution item output code and the change trend item output code, and the clustering of the driving behavior primitive is realized based on an LDA topic model.

[0012] Compared with the prior art, the present application has the following beneficial effects: the sensor collected driving data is filtered by using a sliding average value, sensor noise can be effectively eliminated, data reliability is improved, the driving behavior data is divided from a data structure by using a BMASS method, and thus a driving behavior primitive set P = {P j ;} j＝1:N, can effectively identify the state transition point of the driver during driving, facilitate the analysis of the complex driving behavior of the driver, when the driving behavior primitives are discretized, the data distribution items of the speed and the lateral acceleration are coded based on the comprehensive data percentage characteristics from the experience information, which avoids the problem that the collected driving data distribution is quite different from the standard data set due to the driving habits of the driver, the steering behavior coding of the lateral acceleration and the yaw angle based on the actual road alignment label is beneficial to the accurate identification of the lateral and longitudinal driving behavior of the driver, and the existence of the data trend item can capture the timing of the driving behavior primitives, improve the clustering effect, and avoid clustering errors caused by primitives with similar data distribution, wherein the data trend of the comprehensive data point and the fluctuation of the neighborhood data and the derivative information thereof are obtained, which can effectively avoid misjudgment caused by too frequent data fluctuation, realize accurate judgment of the data trend of the driving behavior primitives, and obtain the discretization result of the driving behavior primitives through the data distribution items and the data trend items of the four variables, and determine the most reasonable driving behavior primitive category number by using the perplexity and consistency score, which can make up for the shortcomings of the LDA model when clustering the time series data, improve the clustering accuracy of the driving behavior primitives, realize accurate expression of the semantic information of the driving behavior primitives, and help the driver accurately and intuitively understand the driving behavior.

[0013] Further, the data distribution item output coding includes speed distribution item output coding, longitudinal acceleration distribution item output coding and steering behavior output coding;

[0014] The speed distribution item output coding includes numbers a, b and c, the a is low speed, the b is medium speed, and the c is high speed;

[0015] The longitudinal acceleration distribution item output coding includes numbers d, e, f, g and h, the d is rapid deceleration, the e is deceleration, the f is constant speed, the g is acceleration, and the h is rapid acceleration;

[0016] The steering behavior output coding is obtained by comprehensive lateral acceleration ay output coding and yaw angle r output coding, the lateral acceleration ay output coding includes numbers s and t, the yaw angle r output coding includes numbers s and t, the s is straight, and the t is turning, when the lateral acceleration ay output coding and the yaw angle r output coding are both turning, the steering behavior output coding outputs turning, otherwise it is identified as straight.

[0017] The beneficial effects of the above step are that: through the data distribution item output coding, the current driving behavior of the driver can be effectively defined, which is convenient for understanding the driving behavior, and the steering behavior coding can be realized by combining the actual road alignment label to accurately divide the lateral and longitudinal driving behavior of the driver.

[0018] Further, the data change trend item output encoding includes a speed change trend item output encoding, a longitudinal acceleration change trend item output encoding, a lateral acceleration change trend item output encoding, and a yaw angle change trend item output encoding.

[0019] The speed change trend item output encoding includes a number vu, vd, and vl.

[0020] The longitudinal acceleration change trend item output encoding includes a number axu, axd, and axl.

[0021] The lateral acceleration change trend item output encoding includes a number ayu, ayd, and ayl.

[0022] The yaw angle change trend item output encoding includes a number ru, rd, and rl.

[0023] The vu is a speed rising trend item output encoding; the vd is a speed falling trend item output encoding; and the vl is a speed stable trend item output encoding.

[0024] The axu is a longitudinal acceleration rising trend item output encoding; the axd is a longitudinal acceleration falling trend item output encoding; and the axl is a longitudinal acceleration stable trend item output encoding.

[0025] The ayu is a lateral acceleration rising trend item output encoding; the ayd is a lateral acceleration falling trend item output encoding; and the ayl is a lateral acceleration stable trend item output encoding.

[0026] The ru is a yaw angle rising trend item output encoding; the rd is a yaw angle falling trend item output encoding; and the rl is a yaw angle stable trend item output encoding.

[0027] The beneficial effects of the above step are that the change trends of the speed, the longitudinal acceleration, the lateral acceleration, and the yaw angle are characterized by the data change trend item output encoding, the time sequence characteristics of the driving behavior primitives are captured, the deficiencies of the LDA topic model when clustering the time sequence are improved, the clustering accuracy of the driving behavior primitives is improved, the semantic information of the driving behavior primitives is accurately expressed, and the driver can accurately and intuitively understand the driving behavior.

[0028] Further, the driving behavior primitive data is preliminarily discretized by preliminarily discretizing the speed v.

[0029] The preliminary discretization of the speed v includes selecting the 20% quantile and the 90% quantile as the discretization thresholds of low speed, medium speed and high speed based on the empirical division information of low speed, medium speed and high speed. The empirical division information generally considers 0-40 km / h as low speed, 40 km / h-70 km / h as medium speed, and above 70 km / h as high speed. The 20% quantile is 11 m / s and the 90% quantile is 20 m / s.

[0030] The beneficial effect of the previous step is that by combining the speed experience classification information, the speed is divided into low speed, medium speed, and high speed. The 20% quantile of 11m / s is selected as the discretization threshold for low speed and medium speed, and the 90% quantile of 20m / s is selected as the discretization threshold for medium speed and high speed. Based on one of the encodings of this output speed distribution item, the speed data is discretized from the data distribution perspective. At the same time, the problem of the collected speed data distribution being significantly different from the standard data set due to the driver's driving habits is avoided, making the judgment of driving behavior more accurate.

[0031] Furthermore, the preliminary discretization of the driving behavior primitive data includes preliminary discretization of the longitudinal acceleration ax;

[0032] The preliminary discretization of the longitudinal acceleration ax includes selecting the 30%, 45%, 60%, and 70% quantiles as the discretization thresholds of sudden deceleration, deceleration, uniform speed, acceleration, and sudden acceleration based on the empirical division information of sudden deceleration, deceleration, uniform speed, acceleration, and sudden acceleration. The empirical division information generally considers that when the absolute value of the acceleration is less than 0.05m / s 2 When the acceleration is too great, the driver cannot sense the presence of acceleration. 2 It will cause physical and mental discomfort to the driver. The 30%, 45%, 60%, and 70% percentiles of the longitudinal acceleration are -0.18 m / s 2 、-0.05m / s 2 , 0.07m / s 2 , 0.18m / s 2 .

[0033] The beneficial effect of the previous step is that by combining the longitudinal acceleration experience division information, the 30% fraction of -0.18m / s is selected. 2 As the discretization threshold for rapid deceleration and deceleration, -0.05m / s at the 45% fraction is selected. 2 As the discretization threshold for deceleration and uniform speed, 0.07m / s at the 60% fraction is selected. 2 As the discretization threshold for uniform speed and acceleration, the 70% quantile of 0.18m / s is selected. 2For the acceleration and rapid acceleration discretization threshold, one of the encodings is output based on the output, which realizes the discretization of longitudinal acceleration data from the perspective of data distribution, avoids the problem that the distribution of collected longitudinal acceleration data is greatly different from the standard data set due to the driving habits of drivers, and makes the judgment of driving behavior more accurate.

[0034] Further, the preliminary discretization of the driving behavior primitive data includes preliminary discretization of the lateral acceleration ay and the yaw angle r; the preliminary discretization of the lateral acceleration ay and the yaw angle r includes determining a straight turning discretization threshold based on a maximum road line shape identification accuracy, the actual road line shape label shows that the straight road segment accounts for 62.80%, and the turning road segment accounts for 37.20%; the maximum line shape identification accuracy is 83.26%, and the lateral acceleration straight turning discrimination threshold corresponding to the maximum line shape identification accuracy is 0.35 m / s 2 , the straight road segment identification accuracy is 90.10%, the turning road segment identification accuracy is 71.72%, and the ay<-0.35 m / s 2 , ay>0.35 m / s 2 is straight, and the ay<-0.35 m / s 2 , ay>0.35 m / s 2 is turning.

[0035] The maximum line shape identification accuracy is 83.33%, and the yaw angle fluctuation straight turning discrimination threshold corresponding to the maximum line shape identification accuracy is 0.04 rad, the straight road segment identification accuracy is 88.80%, and the turning road segment identification accuracy is 74.08%, the yaw angle data change range per second is ≤0.04 rad, and the yaw angle data change range per second is >0.04 rad.

[0036] The beneficial effects of the above step are that: by using the road line shape discrimination threshold of 0.35 m / s 2 with a maximum line shape identification accuracy of 83.26% for the lateral acceleration, as the judgment standard for straight and turning, the straight and turning output encoding of the lateral acceleration is s, t; by using the road line shape discrimination threshold of 0.04 rad with a maximum line shape identification accuracy of 83.33% for the yaw angle r, as the judgment standard for straight and turning, the straight and turning output encoding of the yaw angle is s, t; when the line shape encodings of the lateral acceleration and the yaw angle are both t, it is identified as turning, otherwise it is identified as straight, and the encoding of the turning behavior of the lateral acceleration and the yaw angle based on the actual road line shape label realizes accurate division of the driving behavior of the driver in the lateral and longitudinal directions.

[0037] Further, the data change trend judgment of the driving behavior includes data change trend judgment of the X variable at the t point; the data change trend judgment process of the X variable at the t point is as follows: the X variable is any one of the four variables of speed v, longitudinal acceleration ax, lateral acceleration ay, and yaw angle r;

[0038] The data change trend judgment at the t point based on the derivative of the t point and its neighborhood and the data fluctuation condition includes:

[0039] Step one: obtain the data point x t The variable data SW_X in the neighborhood t And the variable derivative data SW_dX t , the neighborhood range is [max(t-30, 1), min(t+30, NN j )];The NN j is the current driving behavior primitive data length;

[0040] Step two: if max(SW_X t )-min(SW_X t )>α·(max(X)-min(X)), α is a threshold setting ratio, and the value is 0.02, then the data change trend at x t-1 is judged based on the derivative signs of the data points x t , x t+1 , and x t ;

[0041] Step three: if max(SW_X t )-min(SW_X t )≤α·(max(X)-min(X)), then the data change trend at x t is judged based on the sign distribution of SW_dX t .

[0042] The beneficial effects of the above step are that: by setting the neighborhood radius 30, the neighborhood range [max(t-30, 1), min(t+30, NN j )] is selected, the variable data SW_X t in the neighborhood range, the variable derivative data SW_dX t are obtained, the data change trend at the t point is judged, and the process of obtaining the rising, falling, or stable trend considers the time characteristics of the driving behavior primitive sequence, and avoids false judgment caused by local noise;

[0043] When the range of the variable data SW_X t of a certain data point exceeds the set data fluctuation threshold, the data change trend at the t point is judged based on x t itself and the data x t at the previous time point of xt-1 and the data at the next time point x t+1 The derivative of x t Data change trends;

[0044] When a data point corresponds to the variable data SW_X t When the range is within the set data fluctuation threshold or is equal to the threshold range, based on SW_dX t The symbol distribution of x t The data change trend at the location.

[0045] Furthermore, the max(SW_X t )-min(SW_X t )>α·(max(X)-min(X)), if |dx t |>α·min(|max(dX)|,|min(dX)|), then based on dx t The symbol output data point x t The changing trend (up, down), when dx t When the value of the data point x is greater than the derivative judgment threshold, t The trend of change is rising, otherwise the data point x t The changing trend is downward;

[0046] If |dx t |≤α·min(|max(dX)|, |min(dX)|), then based on dx t-1 The symbol output data point x t The changing trend (increasing, decreasing, stable), when dx t-1 When the absolute value of dx is less than or equal to the derivative discrimination threshold, it is stable. t-1 If the derivative threshold is greater than the threshold, it is rising, otherwise it is falling. In particular, when t=1, based on dx t+1 The symbol output data point x t The changing trend (increasing, decreasing, stable), when dx t+1 When the absolute value of dx is less than or equal to the derivative discrimination threshold, it is stable. t+1 If the value is greater than the derivative discrimination threshold, it is rising, otherwise it is falling.

[0047] The beneficial effect of the previous step is: by max(SW_X t )-min(SW_X t )>α·(max(X)-min(X)), dx t The absolute value of dx is outside the threshold range, so t The sign of dx determines the changing trend of the data at point t. tthe absolute value of the derivative of x t-1 or x t+1 The derivative at the point t and the derivative threshold value determine the trend of the data at the point t, and by setting the derivative threshold value, the misjudgment phenomenon caused by data noise is effectively avoided.

[0048] Further, the max(SW_X t )-min(SW_X t )≤α·(max(X)-min(X)), if the number of data points in SW_dX t between [-α·min(|max(dX)|, |min(dX)|) and α·min(|max(dX)|, |min(dX)|) is l, the number of data points greater than α·min(|max(dX)|, |min(dX)|) is m, and the number of data points less than -α·min(|max(dX)|, |min(dX)|) is n.

[0049] If , the trend of the data point x t is rising, if , the trend of the data point x t is falling, otherwise the trend of the data point x t is stable.

[0050] The beneficial effect of the previous step is that if more than three-quarters of the derivative values of the data point corresponding to the variable exceed the derivative threshold value, the trend of the data point x t is rising, if more than three-quarters of the derivative values of the data point exist and are less than the negative derivative threshold value, the trend of the data point x t is falling, otherwise the trend of the data x t is stable. From the distribution of the derivative data, the trend of the data point is determined, and the misjudgment phenomenon caused by data noise is effectively avoided.

[0051] Further, the discretization result is the data distribution item and the data trend item of the speed, the longitudinal acceleration, the lateral acceleration and the yaw angle;

[0052] The discretization result is one of the speed distribution item output encoding, one of the speed trend item output encoding, one of the longitudinal acceleration distribution item output encoding, one of the longitudinal acceleration trend item output encoding, one of the steering behavior output encoding, one of the lateral acceleration trend item output encoding, and one of the yaw angle trend item output encoding.

[0053] The beneficial effects of the previous step are that: the discretization results of the driving behavior primitives are obtained through the data distribution items and the data change trend items of the speed, the longitudinal acceleration, the lateral acceleration and the yaw angle data, the matching of the time series data and the LDA topic model can be realized, the deficiencies of the LDA model in clustering the time series data caused by the bag-of-words model are made up, the clustering accuracy of the driving behavior primitives is improved, the semantic information of the driving behavior primitives is accurately expressed, and the driver can accurately and intuitively understand the driving behavior. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 The data distribution item encoding table of the driving behavior primitive discretization is generated for the discretization thresholds of the speed, the longitudinal acceleration, the lateral acceleration and the yaw angle;

[0055] Figure 2 The flowchart of the data change trend item is generated for the present application;

[0056] Figure 3 The LDA clustering results with or without the data change trend item-entropy and topic difference;

[0057] Figure 4 The LDA clustering results with or without the data change trend item-topic difference color block diagram;

[0058] Figure 5 The LDA clustering results with or without the data change trend item are visualized. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0060] Embodiment 1:

[0061] The embodiment provides a driving behavior primitive clustering method considering data change trend, and the method comprises the following steps: step one: collecting natural driving data of M drivers, selecting the longitudinal driving behavior parameters of speed v and longitudinal acceleration ax, the lateral driving behavior parameters of lateral acceleration ay and yaw angle r as driving behavior analysis variables, performing filtering processing on the driving behavior analysis variable data by using a sliding average value to obtain preprocessed driving behavior data; deriving the preprocessed driving behavior data to obtain driving behavior derivative data; obtaining the driving behavior data set of the M drivers through the preprocessed driving behavior data and the driving behavior derivative data The driving behavior data of the i-th driver, is the driving behavior data of the i-th driver, NM i is the data length of the ith driver; the preprocessed driving behavior data includes speed v, longitudinal acceleration ax, lateral acceleration ay, and yaw angle r, and the driving behavior derivative data includes speed v derivative dv, longitudinal acceleration ax derivative dax, lateral acceleration ay derivative day, and yaw angle r derivative dr;

[0062] Step two: based on the preprocessed driving behavior data, using the BMASS method to process the driving behavior data set to extract driving behavior primitives and obtain a driving behavior primitive set P = {P j ;} j＝1:N N is the total number of primitives; is the jth driving behavior primitive, and NN j is the data length of the jth driving behavior primitive;

[0063] Step three: construct a semantic space, preliminarily discretize the driving behavior primitive data, and obtain a data distribution item output code of the driving behavior primitive;

[0064] The data distribution item output code includes a speed distribution item output code, a longitudinal acceleration distribution item output code, and a steering behavior output code; the speed distribution item output code includes numbers a, b, and c, the a is low speed, the b is medium speed, and the c is high speed; the longitudinal acceleration distribution item output code includes numbers d, e, f, g, and h, the d is rapid deceleration, the e is deceleration, the f is constant speed, the g is acceleration, and the h is rapid acceleration; the steering behavior output code is obtained by combining the lateral acceleration ay output code and the yaw angle r output code, the lateral acceleration ay output code includes numbers s and t, the yaw angle r output code includes numbers s and t, the s is straight, and the t is turning; when the lateral acceleration ay output code and the yaw angle r output code are both turning, the steering behavior output code outputs turning, otherwise it is recognized as straight; the data distribution item output code can effectively define the current driving behavior of the driver, facilitate understanding of the driving behavior, and simultaneously based on the actual road alignment label output steering behavior code can accurately divide the longitudinal and lateral driving behavior of the driver;

[0065] The preliminary discretization of the driving behavior primitive data includes preliminary discretization of the speed v; the preliminary discretization of the speed v includes selecting 20% quantile and 90% quantile as the discretization threshold of low speed, medium speed and high speed according to the speed experience classification information, the speed experience classification information generally considers 0-40 km / h as low speed, 40 km / h-70 km / h as medium speed, and above 70 km / h as high speed, the 20% quantile is 11 m / s, and the 90% quantile is 20 m / s, the speed is divided into low speed, medium speed and high speed by combining the speed experience classification information, 11 m / s of the 20% quantile is selected as the discretization threshold of low speed and medium speed, and 20 m / s of the 90% quantile is selected as the discretization threshold of medium speed and high speed, and one of the speed distribution items is output based on this, the discretization of the speed data is realized from the data distribution angle, and meanwhile, problems such as large difference between the collected speed data distribution and the standard data set caused by the driving habit of the driver are avoided, so that the judgment of the driving behavior is more accurate.

[0066] The preliminary discretization of the driving behavior primitive data includes preliminary discretization of the longitudinal acceleration ax; the preliminary discretization of the longitudinal acceleration ax includes selecting 30%, 45%, 60% and 70% quantiles as the discretization threshold of rapid deceleration, deceleration, constant speed, acceleration and rapid acceleration according to the experience classification information of rapid deceleration, deceleration, constant speed, acceleration and rapid acceleration, the experience classification information generally considers that when the absolute value of acceleration is less than 0.05 m / s 2 , the driver cannot perceive the existence of acceleration, and when the acceleration is greater than 0.2 m / s 2 , the driver will feel uncomfortable in body and mind, the 30%, 45%, 60% and 70% quantiles of the longitudinal acceleration are-0.18 m / s 2 , -0.05 m / s 2 , 0.07 m / s 2 and 0.18 m / s 2 respectively, -0.18 m / s 2 of the 30% quantile is selected as the discretization threshold of rapid deceleration and deceleration, -0.05 m / s 2 of the 45% quantile is selected as the discretization threshold of deceleration and constant speed, 0.07 m / s 2 of the 60% quantile is selected as the discretization threshold of constant speed and acceleration, and 0.18 m / s 2 of the 70% quantile is selected as the discretization threshold of acceleration and rapid acceleration, one of the longitudinal acceleration data is output based on this, the discretization of the longitudinal acceleration data is realized from the data distribution angle, and meanwhile, problems such as large difference between the collected longitudinal acceleration data distribution and the standard data set caused by the driving habit of the driver are avoided, so that the judgment of the driving behavior is more accurate.

[0067] The preliminary discretization of the driving behavior primitive data includes preliminary discretization of lateral acceleration ay and yaw angle r; the preliminary discretization of lateral acceleration ay and yaw angle r includes determining a straight turning discretization threshold based on actual road line shape labels to maximize road line shape identification accuracy, the actual road line shape labels show that straight road segments account for 62.80%, and turning road segments account for 37.20%; the maximum line shape identification accuracy is 83.26%, and the corresponding lateral acceleration straight turning discrimination threshold is 0.35 m / s 2 , the straight road segment identification accuracy is 90.10%, the turning road segment identification accuracy is 71.72%, and the -0.35 m / s 2 ≤ ay ≤ 0.35 m / s 2 is straight, the ay < -0.35 m / s 2 , ay > 0.35 m / s 2 is turning; the maximum line shape identification accuracy is 83.33%, and the corresponding yaw angle fluctuation straight turning discrimination threshold is 0.04 rad, the straight road segment identification accuracy is 88.80%, and the turning road segment identification accuracy is 74.08%, the yaw angle data change range per second ≤ 0.04 rad is straight, and the yaw angle data change range per second > 0.04 rad is turning; when both the lateral acceleration ay and the yaw angle r are in the turning discrimination threshold, it is identified as turning, otherwise it is identified as straight; the road line shape discrimination threshold of the lateral acceleration with the maximum line shape identification accuracy of 83.26% is 0.35 m / s 2 , as the judgment standard of straight and turning, the output code of lateral acceleration straight and turning is s, t; the road line shape discrimination threshold of the yaw angle r change range per second with the maximum line shape identification accuracy of 83.33% is 0.04 rad, as the judgment standard of straight and turning, the output code of yaw angle straight and turning is s, t; when the line shape codes of the two parameters of lateral acceleration and yaw angle are both t, it is identified as turning, otherwise it is identified as straight. The coding of turning behavior based on lateral acceleration and yaw angle realizes accurate division of the driver's lateral and longitudinal driving behavior.

[0068] Step four: based on the derivative, data fluctuation of the t point and its neighborhood in the driving behavior primitive P j , the data change trend of the driving behavior primitive is judged to obtain the data change trend item output code of the driving behavior primitive;

[0069] The data at the t point includes four driving behavior variables: speed v, longitudinal acceleration ax, lateral acceleration ay, and yaw angle r;

[0070] The data change trend item output encoding includes a speed change trend item output encoding, a longitudinal acceleration change trend item output encoding, a lateral acceleration change trend item output encoding, and a yaw angle change trend item output encoding.

[0071] The speed change trend item output encoding includes numbers vu, vd, and vl.

[0072] The longitudinal acceleration change trend item output encoding includes numbers axu, axd, and axl.

[0073] The lateral acceleration change trend item output encoding includes numbers ayu, ayd, and ayl.

[0074] The yaw angle change trend item output encoding includes numbers ru, rd, and rl.

[0075] The vu is a speed increase trend item output encoding; the vd is a speed decrease trend item output encoding; and the vl is a speed stable trend item output encoding.

[0076] The axu is a longitudinal acceleration increase trend item output encoding; the axd is a longitudinal acceleration decrease trend item output encoding; and the axl is a longitudinal acceleration stable trend item output encoding.

[0077] The ayu is a lateral acceleration increase trend item output encoding; the ayd is a lateral acceleration decrease trend item output encoding; and the ayl is a lateral acceleration stable trend item output encoding.

[0078] The ru is a yaw angle increase trend item output encoding; the rd is a yaw angle decrease trend item output encoding; and the rl is a yaw angle stable trend item output encoding.

[0079] By characterizing the change trends of the speed, the longitudinal acceleration, the lateral acceleration, and the yaw angle through the data change trend item output encoding, the timing characteristics of the driving behavior primitives can be captured, the shortcomings of the LDA topic model when clustering the time series can be improved, the clustering accuracy of the driving behavior primitives can be improved, the semantic information of the driving behavior primitives can be accurately expressed, and the driver can accurately and intuitively understand the driving behavior.

[0080] Step five: integrating the data distribution item output encoding and the data change trend item output encoding of the driving behavior primitives to obtain a discretization result of the driving behavior primitives, the discretization result including the distribution item output encoding and the change trend item output encoding, and clustering the driving behavior primitives based on the LDA topic model.

[0081] The driving data collected by the sensor is filtered by the sliding average value, which can effectively eliminate sensor noise and improve data reliability. The driving behavior data is divided using the BMASS method from the data structure, so as to obtain a driving behavior primitive set P={P j j＝1:N , which can effectively identify the state transition point of the driver during driving, facilitate the analysis of the complex driving behavior of the driver, and avoid the problem that the collected driving data distribution is greatly different from the standard data set due to the driving habits of the driver. The steering behavior coding of the lateral acceleration and the yaw angle based on the actual road alignment label is beneficial to the accurate identification of the lateral and longitudinal driving behavior of the driver. The existence of the data change trend item can capture the time sequence of the driving behavior primitive, improve the clustering effect, and avoid clustering errors caused by primitives with similar data distribution. The data point and its neighborhood data fluctuation and derivative information are used to obtain the data change trend of the data point, which can effectively avoid misjudgment caused by too frequent data fluctuation, realize accurate judgment of the data change trend of the driving behavior primitive, and obtain the discretization result of the driving behavior primitive through the data distribution item and the data change trend item of the four variables. The use of the perplexity and consistency score to determine the most reasonable driving behavior primitive category number can make up for the shortcomings of the LDA model in clustering time sequence data caused by the bag-of-words model, improve the clustering accuracy of the driving behavior primitive, realize accurate expression of the semantic information of the driving behavior primitive, and help the driver accurately and intuitively understand the driving behavior.

[0082] Further, the data change trend judgment of the driving behavior includes data change trend judgment of the X variable at the t point. The data change trend judgment process of the X variable at the t point is as follows: the X variable is any one of the four variables of speed v, longitudinal acceleration ax, lateral acceleration ay, and yaw angle r. The data change trend of the t point is judged based on the derivative and data fluctuation of the t point and its neighborhood, which includes the following steps: step one: obtaining the variable data SW_X t and the variable derivative data SW_dX t in the neighborhood of the data point x t , and the neighborhood range is [max(t-30, 1), min(t+30, NN j )]; the NN j is the current driving behavior primitive data length;

[0083] Step two: if max(SW_X t )-min(SW_X t ) > a · (max(X)-min(X)), a is a threshold setting ratio, and the value is 0.02, then the data change trend of the t point is judged based on the data point x​t-1 , x t , x t+1 , x t , x t , x t , x t , x t , x j , x t , x t , x t , x t , x t , x t-1 , x t+1 , x t , x t , x t , x t , x , x

[0084] , x t , x t , x t , x t , x t , x t , x t , x t , x t , x t-1 , x t , xt-1 is less than or equal to the derivative threshold value, is stable, when dx t-1 is greater than the derivative threshold value, is rising, otherwise is falling, in particular, when t = 1, the sign of dx t+1 is outputted to indicate the trend of data point x t , when dx t+1 is less than or equal to the derivative threshold value, is stable, when dx t+1 is greater than the derivative threshold value, is rising, otherwise is falling. When max(SW_X t )-min(SW_X t )>a·(max(X)-min(X)), the absolute value of dx t is out of the threshold value range, thus the trend of data at t point is determined based on the sign of dx t , the absolute value of dx t is in the threshold value range, thus the trend of data at t point is determined based on the relationship between the derivative of x t-1 or x t+1 and the derivative threshold value, by setting the derivative threshold value, the misjudgment phenomenon of data trend caused by data noise is effectively avoided.

[0085] max(SW_X t )-min(SW_X t )≤a·(max(X)-min(X)), if the number of data points in SW_dX t between [-a·min(|max(dX)|, |min(dX)|), a·min(|max(dX)|, |min(dX)|)] is denoted as l, the number of data points greater than a·min(|max(dX)|, |min(dX)|) is denoted as m, and the number of data points less than -a·min(|max(dX)|, |min(dX)|) is denoted as n; if , the trend of data point x t is rising, if , the trend of data point x t is falling, otherwise the trend of data point x t is stable. If more than three quarters of the derivative values of the data points corresponding to the variable derivative data are greater than the derivative threshold value, the trend of data point x t is rising, if more than three quarters of the derivative values of the data points are less than the negative derivative threshold value, the trend of data point x t is falling, otherwise data x tThe change trend is stable, the change trend of the data points is discriminated based on the derivative data distribution, and misjudgment caused by data noise is effectively avoided.

[0086] The discretization result is a data distribution item and a data change trend item of the speed, the longitudinal acceleration, the lateral acceleration and the yaw angle; the discretization result is one of speed distribution item output encodings, one of speed change trend item output encodings, one of longitudinal acceleration distribution item output encodings, one of longitudinal acceleration change trend item output encodings, one of steering behavior output encodings, one of lateral acceleration change trend item output encodings, and one of yaw angle change trend item output encodings. The discretization result of the driving behavior primitive is obtained through the data distribution item and the data change trend item of the speed, the longitudinal acceleration, the lateral acceleration and the yaw angle, the matching of the time series data and the LDA topic model is realized, the deficiency of the LDA model in clustering the time series data caused by the bag-of-words model is made up, the clustering accuracy of the driving behavior primitive is improved, the semantic information of the driving behavior primitive is accurately expressed, and the driver can accurately and intuitively understand the driving behavior.

[0087] The above examples are only used to illustrate the technical solutions of the present application, but not limit the same; based on the examples of the present application, all other examples obtained by those skilled in the art without creative work fall within the scope of protection of the present application. Although the present application is described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the examples of the present application.

Claims

1. A driving behavior primitive clustering method considering data variation trend, characterized in that, The method comprises the following steps: Step one: collecting natural driving data of M drivers, selecting longitudinal driving behavior parameters speed v, longitudinal acceleration ax, lateral driving behavior parameters lateral acceleration ay, and yaw angle r as driving behavior analysis variables, using sliding average to filter driving behavior analysis variable data to obtain pre-processed driving behavior data, and deriving the pre-processed driving behavior data to obtain driving behavior derivative data; The driving behavior data set of the Mth driver is obtained by preprocessing the driving behavior data and the driving behavior derivative data , wherein is the driving behavior data of the ith driver, , is the data length of the ith driver, the preprocessed driving behavior data includes speed v, longitudinal acceleration ax, lateral acceleration ay, and yaw angle r, and the driving behavior derivative data includes speed v derivative dv, longitudinal acceleration ax derivative dax, lateral acceleration ay derivative day, and yaw angle r derivative dr. Step two: based on the pre-processed driving behavior data, using BMASS method to process the driving behavior data set to extract driving behavior primitives, and obtain a driving behavior primitive set ; N is the total number of primitives; is the jth driving behavior primitive, is the data length of the jth driving behavior primitive; Step three: constructing a semantic space, preliminarily discretizing the driving behavior primitive data to obtain data distribution item output coding of the driving behavior primitive; Step four: driving behavior primitive based on P j The derivative of the middle point and its neighborhood, the data fluctuation, and the data change trend of the driving behavior primitive are judged to obtain the data change trend item output code of the driving behavior primitive. The data at point t includes four driving behavior variables: speed v, longitudinal acceleration ax, lateral acceleration ay, and yaw angle r; Step five: synthesizing the data distribution item output coding and the data change trend item output coding of the driving behavior primitive to obtain the discretization result of the driving behavior primitive, wherein the discretization result comprises the distribution item output coding and the change trend item output coding, and clustering the driving behavior primitive based on an LDA topic model.

2. The driving behavior primitive clustering method considering data variation trend according to claim 1, characterized in that, The data distribution item output coding comprises speed distribution item output coding, longitudinal acceleration distribution item output coding, and steering behavior output coding; The speed distribution item output coding comprises numbers a, b, and c, wherein the a is low speed, the b is medium speed, and the c is high speed; The longitudinal acceleration distribution item output coding comprises numbers d, e, f, g, and h, wherein the d is rapid deceleration, the e is deceleration, the f is constant speed, the g is acceleration, and the h is rapid acceleration; The steering behavior output coding is obtained by synthesizing the lateral acceleration ay output coding and the yaw angle r output coding, wherein the lateral acceleration ay output coding comprises numbers s and t, the yaw angle r output coding comprises numbers s and t, the s is straight driving, and the t is turning; when the lateral acceleration ay output coding and the yaw angle r output coding are both turning, the steering behavior output coding outputs turning, otherwise, it is identified as straight driving.

3. The driving behavior primitive clustering method considering data variation trend according to claim 1, characterized in that, The data change trend item output coding comprises speed change trend item output coding, longitudinal acceleration change trend item output coding, lateral acceleration change trend item output coding, and yaw angle change trend item output coding; The speed change trend item output coding comprises numbers vu, vd, and vl; The longitudinal acceleration change trend item output coding comprises numbers axu, axd, and axl; The lateral acceleration change trend item output coding comprises numbers ayu, ayd, and ayl; The yaw angle change trend item output coding comprises numbers ru, rd, and rl; The vu is the speed rising trend item output coding; The vd is the speed falling trend item output coding; and the vl is the speed stable trend item output coding; The axu is the longitudinal acceleration rising trend item output coding; The axd is the longitudinal acceleration falling trend item output coding; The axl is the longitudinal acceleration stable trend item output coding; The ayu is the lateral acceleration rising trend item output coding; The ayd is the lateral acceleration falling trend item output coding; The ayl is the lateral acceleration stable trend item output coding; The ru is a yaw angle rising trend item output code; The rd is a yaw angle falling trend item output code; The rl is a yaw angle stable trend item output code.

4. The driving behavior primitive clustering method considering data variation trend according to claim 1, characterized in that, The preliminary discretization of the driving behavior primitive data includes preliminary discretization of the speed v; The preliminary discretization of the speed v includes selecting 20% quantile and 90% quantile as the discretization threshold of low speed, medium speed and high speed according to low speed, medium speed and high speed experience division information, the experience division information considers 0-40 km / h as low speed, 40 km / h-70 km / h as medium speed, and above 70 km / h as high speed, the 20% quantile is 11 m / s, and the 90% quantile is 20 m / s.

5. The driving behavior primitive clustering method considering data variation trend according to claim 1, characterized in that, The preliminary discretization of the driving behavior primitive data includes preliminary discretization of the longitudinal acceleration ax; The preliminary discretization of the longitudinal acceleration ax includes selecting 30%, 45%, 60%, 70% quantile of the longitudinal acceleration ax as the discretization threshold of the sharp deceleration, deceleration, uniform speed, acceleration, sharp acceleration according to the empirical classification information of the sharp deceleration, deceleration, uniform speed, acceleration, sharp acceleration, and the empirical classification information is -0.2 m / s 2 , -0.5 m / s 2 , 0.05 m / s 2 , 0.2 m / s 2 , and the 30%, 45%, 60%, 70% quantile of the longitudinal acceleration is -0.18 m / s 2 , -0.05 m / s 2 , 0.07 m / s 2 , 0.18 m / s 2 .

6. The driving behavior primitive clustering method considering data variation trend according to claim 1, characterized in that, The preliminary discretization of the driving behavior primitive data includes preliminary discretization of lateral acceleration ay and yaw angle r; the preliminary discretization of the lateral acceleration ay and the yaw angle r includes determining a straight turning discretization threshold based on actual road line shape label maximum road line shape identification accuracy, the lateral acceleration straight turning discrimination threshold is 0.35 m / s 2 , the -0.35 m / s 2 ≤ ay≤0.35 m / s 2 is straight, the ay< -0.35 m / s 2 , ay> 0.35 m / s 2 is turning; The straight turning discrimination threshold of the yaw angle fluctuation is 0.04 rad, the data change range of the yaw angle per second is ≤0.04 rad for straight line, and the data change range of the yaw angle per second is >0.04 rad for turning.

7. The driving behavior primitive clustering method considering data variation trend according to claim 1, characterized in that, The data change trend judgment of the driving behavior primitive includes data change trend judgment of the X variable at the t point; the data change trend judgment process of the X variable at the t point is as follows: the X variable is any one of the four variables of speed v, longitudinal acceleration ax, lateral acceleration ay and yaw angle r; The data change trend judgment of the t point based on the derivative of the t point and its neighborhood and the data fluctuation includes: Step one: Acquire data points Variable data within a neighborhood And variable derivative data , neighborhood range is ; the NN j is the current driving behavior primitive data length; Step two: If , The threshold ratio is set to 0.02, and the data point , , The derivative sign at the point is used to determine the trend of data change at the point . Step three: If , then determine the trend of data at based on the symbol distribution of .

8. The driving behavior primitive clustering method considering data variation trend according to claim 7, characterized in that, The If , the change trend of the data point is rising, falling, when is greater than the derivative threshold value, at this time the change trend of the data point is rising, otherwise the change trend of the data point is falling. ​ if , then based on Symbolic output data points The changing trend is rising, falling, and stable. When the absolute value of is less than or equal to the derivative discrimination threshold, it is stable. If it is greater than the derivative judgment threshold, it is rising, otherwise it is falling. When, based on Symbolic output data points The changing trend is rising, falling, and stable. When the absolute value of is less than or equal to the derivative discrimination threshold, it is stable. If the value is greater than the derivative discrimination threshold, it is rising, otherwise it is falling.

9. The driving behavior primitive clustering method considering data variation trend according to claim 7, characterized in that, if the number of data points in the interval between the two points is denoted by l, the number of data points greater than is denoted by m, and the number of data points less than is denoted by n; If then the trend of the data points is increasing, if then the trend of the data points is decreasing, and otherwise the trend of the data points is stable.

10. The driving behavior primitive clustering method considering data variation trend according to claim 1, characterized in that, The discretization result is the data distribution item and the data change trend item of the speed, the longitudinal acceleration, the lateral acceleration and the yaw angle; The discretization result is one of the speed distribution item output code, one of the speed change trend item output code, one of the longitudinal acceleration distribution item output code, one of the longitudinal acceleration change trend item output code, one of the steering behavior output code, one of the lateral acceleration change trend item output code, and one of the yaw angle change trend item output code.

Images