A driving event clustering method and system based on an improved LDA model

Abstract

The application discloses a driving event clustering method and system based on an improved LDA model, and comprises the following steps: acquiring driving events, each driving event comprising a plurality of driving data points, and acquiring the longitudinal jerk of the driving data points; dividing the speed into low speed, medium speed and high speed; based on vehicle longitudinal dynamics, dividing the longitudinal acceleration into slow acceleration, rapid acceleration, uniform speed, slow deceleration and rapid deceleration according to a threshold value; based on the limitation of a tire friction circle, dividing the lateral acceleration into low side shift and high side shift according to a threshold value; dividing the longitudinal jerk into slow jerk acceleration, rapid jerk acceleration, slow jerk deceleration and rapid jerk deceleration; based on the categories of speed, longitudinal acceleration, lateral acceleration and longitudinal jerk, performing discrete processing on the driving data points of the driving events to obtain discrete data groups; inputting the driving events represented by the discrete data groups into an LDA model, setting the number of clusters of the LDA model, and clustering the driving events.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of data processing, and particularly relates to a driving event clustering method and system based on an improved LDA model. BACKGROUND

[0002] Driving events are the most basic driving behavior units and the smallest driving data segments with physical meaning. Analyzing driving behavior is the basis for improving traffic safety and developing intelligent vehicles. Therefore, obtaining driving event semantics efficiently and accurately to describe driving behavior is the focus of current research.

[0003] Due to the similarity between driving data and natural language, a text topic model (LDA) can be used to achieve the purpose of driving event clustering. However, the input requirement of the LDA model is a sample set composed of discrete data points, which makes it difficult to apply in the field. Therefore, scholars have proposed an improved LDA model based on continuous driving data. The continuous driving data points are discretized, and the driving events are clustered based on the discrete data points through the LDA model. This method directly analyzes massive driving data, reduces the limitations of manual calibration, and avoids the complexity of feature engineering and its impact on subsequent results, making it easier and faster for people to accurately understand driving behavior.

[0004] However, the existing method for clustering driving events based on the improved LDA model has some limitations. When discretizing continuous data, the sample data range and data point density are not considered comprehensively. Moreover, the vehicle dynamics are not coupled, which leads to inaccurate classification of discrete data points. For example, continuous data in a small range may be discretized into different categories, or data in a large range may be discretized into the same category, which seriously affects the subsequent clustering results. SUMMARY

[0005] To address the above problems, the present application provides a driving event clustering method based on an improved LDA model, which includes the following steps:

[0006] Obtain a plurality of driving events, each of which includes a plurality of driving data points, each of which includes speed, longitudinal acceleration, and lateral acceleration. Calculate the longitudinal jerk of the driving data points based on the longitudinal acceleration.

[0007] Divide the speed into three categories: low speed, medium speed, and high speed, according to a predetermined speed threshold.

[0008] Based on the longitudinal dynamics of the vehicle, analyze the speed and longitudinal acceleration of the driving data points, obtain the threshold value of the longitudinal acceleration at different speeds, and divide the longitudinal acceleration into five categories: slow acceleration, rapid acceleration, constant speed, slow deceleration, and rapid deceleration.

[0009] Based on the restriction of the tire friction circle, the longitudinal acceleration and the lateral acceleration of the driving data points are coupled and analyzed, the threshold value of the lateral acceleration under different longitudinal acceleration is obtained, and the lateral acceleration is divided into two categories of low side shift and high side shift according to the threshold value;

[0010] According to the preset longitudinal jerk threshold value, the longitudinal jerk is divided into four categories of slow acceleration, rapid acceleration, slow deceleration and rapid deceleration;

[0011] Based on the speed category, the longitudinal acceleration category, the lateral acceleration category and the jerk category, the driving data points of the driving event are discretely processed to obtain a discrete data group;

[0012] The driving event represented by the discrete data group is input into the LDA model, the cluster number of the LDA model is set, and the driving event is clustered.

[0013] Compared with the prior art, the beneficial effects of the present application are: the technical scheme provides a new data discretization method, which combines the discretization of data with the LDA model through coupling analysis, obtains an improved LDA model, and has better discretization effect, thereby facilitating to obtain more accurate clustering results.

[0014] Preferably, the original data points are smoothed by the moving average method to obtain the driving data points;

[0015] The longitudinal jerk of the driving data points is obtained, and the longitudinal jerk is smoothed by the moving average method.

[0016] The beneficial effects of the preferred scheme are: the collected original data points may have abnormal values and noise, etc., and through smoothing, the noise of the original data points and the longitudinal jerk can be removed to obtain effective driving data points, which is beneficial to reduce the discretization error.

[0017] Preferably, 30km / h and 60km / h are set as the speed threshold value, and the speed is v,

[0018] If 0km / h≤v≤30km / h, the speed category is low speed;

[0019] If 30km / h<v≤60km / h, the speed category is medium speed;

[0020] If v>60km / h, the speed category is high speed.

[0021] Preferably, the speed v is taken as the independent variable, the longitudinal acceleration a x is taken as the dependent variable, and the v-a x scatter plot of the driving data points is output.

[0022] Obtain the maximum longitudinal acceleration a corresponding to each velocity v. x1 and minimum longitudinal acceleration a x2 The fitting yielded va x Coupled shape;

[0023] Based on va x The variation pattern of the coupling shape is used to determine va. x The sign of the slope of the upper envelope and the sign of the slope of the lower envelope in a scatter plot;

[0024] Based on the sign of the slope of the upper envelope, obtain the value va located on the upper envelope. x The va is calculated based on the coordinates of the two peak points of the scatter plot. x Upper envelope: a x,max =f1(v);

[0025] Based on the sign of the slope of the lower envelope, obtain the value va located on the lower envelope. x The va is calculated based on the coordinates of the two peak points of the scatter plot. x Lower envelope: a x,min =f2(v).

[0026] The advantages of this preferred solution are: first, determine the sign of the slope of the upper and lower envelopes, and then select the peak point on the coupling shape based on the sign; otherwise, it is easy to cause a large error in the envelope.

[0027] Preferably, the longitudinal acceleration a is set. x The uniform velocity range is -0.05 m / s. 2 ~0.05m / s 2 ,

[0028] If a x >0.05m / s 2 Then the longitudinal acceleration a x The categories are gradual acceleration or rapid acceleration;

[0029] If a x <-0.05m / s 2 Then the longitudinal acceleration a x The categories are gradual deceleration or rapid deceleration;

[0030] Obtain the upper envelope a x,max =a corresponding to velocity v on f1(v) x With 0.05m / s 2 The difference And Multiply by 50% to get: Fitting speed v and Based on the dependency relationship, the thresholds for gradual acceleration and rapid acceleration are obtained:

[0031] like Then the longitudinal acceleration a x The category is slow acceleration;

[0032] like Then the longitudinal acceleration a x The category is rapid acceleration;

[0033] Obtain the lower envelope a x,min =a corresponding to the velocity v on f2(v) x With -0.05m / s 2 The difference And Multiply by 50% to get: Fitting speed v and Based on the dependency relationship, the thresholds for gradual acceleration and rapid acceleration are obtained:

[0034] like Then the longitudinal acceleration a x The category is slowing down;

[0035] like Then the longitudinal acceleration a x The category is rapid deceleration.

[0036] The beneficial effects of this preferred solution are as follows: by calculating the thresholds for gradual acceleration, rapid acceleration, gradual deceleration, and rapid deceleration using the above method, longitudinal acceleration is divided into five categories, which helps to improve the accuracy of driving data discretization.

[0037] Preferably, based on the limitation of the tire friction circle, the longitudinal acceleration a x Let a be the independent variable and the lateral acceleration. y As the dependent variable, output 'a' for the driving data points. x -a y Scatter plot;

[0038] Get the point (a) on the scatter plot that is farthest from the origin (0, 0). x0 a y0 ), with the origin of the scatter plot as the center, and (a x0 a y0 The distance between the origin and the point is a, with radius r, to obtain a. x -a y Fitting circle for coupled shape: a y 2 =r 2 -a x 2 Let the fitted circle be a. x-a y envelope line of a

[0039] with the origin of the scatter plot as the center and as the radius, to obtain a fitting circle representing the lateral acceleration threshold:

[0040] If a y 2 ≤ a y,50％ 2 , the category of the lateral acceleration a y is low side shift;

[0041] If a y 2 > a y,50％ 2 , the category of the lateral acceleration a y is high side shift.

[0042] The beneficial effect of the preferred scheme is that when determining whether the lateral acceleration is low side shift or high side shift, the longitudinal acceleration value at the same time as the lateral acceleration needs to be considered comprehensively, the longitudinal acceleration and the lateral acceleration are coupled and analyzed through the limitation of the tire friction circle, and the accuracy of classifying the lateral acceleration is improved.

[0043] Preferably, the derivative of the longitudinal acceleration is obtained, the longitudinal jerk is obtained based on the derivative, and 3.6 m / s 3 and -3.6 m / s 3 are set as the longitudinal jerk threshold, and the longitudinal jerk is j.

[0044] If 0 m / s 3 ≤ j ≤ 3.6 m / s 3 , the category of the longitudinal jerk j is slow jerk acceleration;

[0045] If j > 3.6 m / s 3 , the category of the longitudinal jerk j is rapid jerk acceleration;

[0046] If -3.6 m / s 3 ≤ j < 0 m / s 3 , the category of the longitudinal jerk j is slow jerk deceleration;

[0047] If j < -3.6 m / s 3 , the category of the longitudinal jerk j is rapid jerk deceleration.

[0048] The beneficial effect of the preferred scheme is that according to the actual driving conditions, 3.6 m / s 3 and -3.6 m / s 3 are set as the longitudinal jerk threshold, and the longitudinal jerk is classified.

[0049] Preferably, the driving data points are discretized based on the speed category, the longitudinal acceleration category, the lateral acceleration category and the longitudinal jerk category to obtain a discrete data set;

[0050] The driving event characterized by the discrete data set is input into the LDA model, and the number of clusters of the LDA model is set to cluster the driving events.

[0051] The beneficial effect of the preferred embodiment is that the continuous driving data points are discretized, which is equivalent to classifying the driving data points, for example, 100 driving data points are discretized into 50 discrete data sets, and the LDA model clusters the driving events characterized by the 50 discrete data sets.

[0052] In a second aspect, the application designs a driving event clustering system based on an improved LDA model, comprising:

[0053] The driving event acquisition module is configured to acquire a plurality of driving events, each of which includes a plurality of driving data points, and each of the driving data points includes speed, longitudinal acceleration and lateral acceleration;

[0054] The longitudinal jerk output module is configured to calculate the longitudinal jerk of the driving data points based on the longitudinal acceleration;

[0055] The speed category division module is configured to divide the speed into three categories, low speed, medium speed and high speed, according to a preset speed threshold;

[0056] The longitudinal acceleration category division module is configured to analyze the speed and longitudinal acceleration of the driving data points based on vehicle longitudinal dynamics, obtain the threshold value of the longitudinal acceleration under different speeds, and divide the longitudinal acceleration into five categories, slow acceleration, rapid acceleration, constant speed, slow deceleration and rapid deceleration, according to the threshold value;

[0057] The lateral acceleration category division module is configured to analyze the longitudinal acceleration and lateral acceleration of the driving data points based on the limitation of the tire friction circle, obtain the threshold value of the lateral acceleration under different longitudinal accelerations, and divide the lateral acceleration into two categories, low side shift and high side shift, according to the threshold value;

[0058] The longitudinal jerk category division module is configured to divide the longitudinal jerk into four categories, slow jerk acceleration, rapid jerk acceleration, slow jerk deceleration and rapid jerk deceleration, according to a preset longitudinal jerk threshold;

[0059] The driving data point discretization module is configured to discretize driving data points based on the categories of speed, longitudinal acceleration, lateral acceleration and longitudinal jerk to obtain a discrete data group.

[0060] The clustering module is configured to input the driving events represented by the discrete data group numbers into the LDA model, set the number of clusters for the LDA model, and cluster the driving events.

[0061] Compared with the prior art, the system provides a new data discretization method, combines data discretization with an LDA model, obtains an improved LDA model, has better discretization effect, and is thus beneficial to obtaining more accurate clustering results. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 A flowchart of the driving event clustering method of the embodiment of the present application;

[0063] Figure 2 A threshold value diagram for dividing the categories of longitudinal acceleration;

[0064] Figure 3 A threshold value diagram for dividing the categories of lateral acceleration;

[0065] Figure 4 A curve diagram of the perplexity varying with the number of clusters. DETAILED DESCRIPTION

[0066] To make the objectives, technical solutions and advantages of the present application clearer, the technical solutions in the embodiments of the present application will be described below in a clear and complete manner with reference to the embodiments of the present application.

[0067] Embodiment 1

[0068] The embodiment provides a driving event clustering method based on an improved LDA model, including the following steps:

[0069] Step 1: Obtain a plurality of driving events, each of which includes a plurality of driving data points, each of which includes speed, longitudinal acceleration and lateral acceleration. Preferably, the driving events include a plurality of original data points. Since the original data points may have problems such as outliers, missing values, noise, etc., the original data points are smoothed according to the moving average method to obtain effective driving data points. The longitudinal jerk of the driving data points is calculated by derivation, and the effective longitudinal jerk is obtained by smoothing the longitudinal jerk according to the moving average method.

[0070] The driving event can be one or more, such as driving event A and driving event B in total, event A includes 10 driving data points, and event B includes 15 driving data points. Clustering driving events is clustering event A and event B represented by the 25 data points.

[0071] Step 2: According to the preset speed threshold, the speed is divided into three categories: low speed, medium speed and high speed. In this embodiment, 30 km / h and 60 km / h are set as the speed threshold, and the speed is v,

[0072] If 0 km / h≤v≤30 km / h, the speed category is low speed;

[0073] If 30 km / h<v≤60 km / h, the speed category is medium speed;

[0074] If v>60 km / h, the speed category is high speed.

[0075] Step 3: Based on the longitudinal dynamics of the vehicle, the speed and longitudinal acceleration of the driving data points are coupled and analyzed, taking the speed v as the independent variable, the longitudinal acceleration a x as the dependent variable, and outputting the v-a x scatter plot of the driving data points;

[0076] Obtain the maximum longitudinal acceleration a x1 and the minimum longitudinal acceleration a x2 corresponding to each speed v, and fit to obtain the v-a x coupling shape; based on the change rule of the v-a x coupling shape, determine the positive and negative of the slope of the upper envelope line of the v-a x scatter plot and the positive and negative of the slope of the lower envelope line;

[0077] According to the positive and negative of the slope of the upper envelope line, obtain two peak points of the v-a x scatter plot on the upper envelope line, and based on the coordinates of the two peak points, calculate to obtain the upper envelope line of the v-a x : a x,max =f1(v);

[0078] Based on the sign of the slope of the lower envelope, obtain the value va located on the lower envelope. x The va is calculated based on the coordinates of the two peak points of the scatter plot. x Lower envelope: a x,min =f2(v).

[0079] Set longitudinal acceleration a x The uniform velocity range is -0.05 m / s. 2 ~0.05m / s 2 ,

[0080] If a x >0.05m / s 2 Then the longitudinal acceleration a x The categories are gradual acceleration or rapid acceleration;

[0081] If a x <-0.05m / s 2 Then the longitudinal acceleration a x The categories are gradual deceleration or rapid deceleration.

[0082] Obtain the upper envelope a x,max =a corresponding to velocity v on f1(v) x With 0.05m / s 2 The difference And Multiply by 50% to get: Fitting speed v and Based on the dependency relationship, the thresholds for gradual acceleration and rapid acceleration are obtained:

[0083] like Then the longitudinal acceleration a x The category is slow acceleration;

[0084] like Then the longitudinal acceleration a x The category is rapid acceleration;

[0085] Obtain the lower envelope a x,min =a corresponding to the velocity v on f2(v) x With -0.05m / s 2 The difference And Multiply by 50% to get: Fitting speed v and Based on the dependency relationship, the thresholds for gradual acceleration and rapid acceleration are obtained:

[0086] like Then the longitudinal acceleration ax The category of the longitudinal acceleration a

[0087] If the longitudinal acceleration a x The category of the longitudinal acceleration a

[0088] Step 4: Based on the limit of the tire friction circle, taking the longitudinal acceleration a x as the independent variable, the lateral acceleration a y as the dependent variable, output the a x -a y scatter plot of the driving data points;

[0089] Get the point (a x0 , a y0 ) farthest from the origin (0, 0) on the scatter plot, take the origin of the scatter plot as the center, and the distance between (a x0 , a y0 ) and the origin as the radius r, get the fitting circle of the a x -a y coupling shape: a y 2 = r 2 -a x 2 Set the fitting circle as the envelope of a x -a y ;

[0090] Take the origin of the scatter plot as the center, and as the radius, get the fitting circle representing the lateral acceleration threshold:

[0091] If a y 2 ≤ a y,50％ 2 , the lateral acceleration a y The category of the lateral acceleration a

[0092] If a y 2 > a y,50％ 2 , the lateral acceleration a y The category of the lateral acceleration a

[0093] Step 5: Get the derivative of the longitudinal acceleration, get the longitudinal jerk based on the derivative, set 3.6 m / s 3 and -3.6 m / s 3 as the longitudinal jerk threshold, and set the longitudinal jerk as j

[0094] If 0 m / s 3 ≤ j ≤ 3.6 m / s 3If the longitudinal jerk j is , then the category is gradual acceleration;

[0095] If j > 3.6 m / s 3 If the longitudinal jerk j is , then the category is rapid acceleration;

[0096] If -3.6m / s 3 ≤j<0m / s 3 If the longitudinal jerk j is , then the category is deceleration acceleration;

[0097] If j < -3.6 m / s 3 If the longitudinal jerk j is , then the category is rapid deceleration.

[0098] Step 6: Discretize the driving data points based on the categories of speed, longitudinal acceleration, lateral acceleration, and longitudinal jerk, to obtain discrete data sets. The discrete data sets are combinations of various categories. For example, low speed + gentle acceleration + high lateral movement + rapid acceleration is one type of discrete data set, and medium speed + gentle deceleration + high lateral movement + gentle acceleration is another type. The specific number of discrete data sets is determined by the discretization results of the driving data points.

[0099] Driving events represented by discrete data sets are input into the LDA model, the number of clusters is set for the LDA model, and the driving events are clustered.

[0100] This embodiment also provides a driving event clustering system based on an improved LDA model, including:

[0101] A driving event acquisition module is configured to acquire a number of driving events, each driving event including a number of driving data points, each driving data point including speed, longitudinal acceleration and lateral acceleration;

[0102] A longitudinal acceleration output module is configured to calculate the longitudinal acceleration of the driving data points based on the longitudinal acceleration.

[0103] The speed category classification module is configured to classify speeds into three categories: low speed, medium speed, and high speed, based on preset speed thresholds.

[0104] The longitudinal acceleration classification module is configured to perform coupled analysis of the speed and longitudinal acceleration of driving data points based on vehicle longitudinal dynamics, obtain the threshold of longitudinal acceleration at different speeds, and classify the longitudinal acceleration into five categories: slow acceleration, rapid acceleration, constant speed, slow deceleration, and rapid deceleration according to the threshold.

[0105] The lateral acceleration category division module is configured to perform a coupled analysis on the longitudinal acceleration and lateral acceleration of driving data points based on the limitation of the tire friction circle, obtain the thresholds of the lateral acceleration under different longitudinal accelerations, and divide the lateral acceleration into two categories, namely low lateral shift and high lateral shift, according to the thresholds;

[0106] The longitudinal jerk category division module is configured to divide the longitudinal jerk into four categories, namely slow acceleration, rapid acceleration, slow deceleration, and rapid deceleration, according to a preset longitudinal jerk threshold;

[0107] The driving data point discretization module is configured to perform discretization processing on driving data points based on the categories of speed, longitudinal acceleration, lateral acceleration, and longitudinal jerk, and obtain a discrete data group; the driving data point discretization module includes: a discrete data group editing unit configured to obtain the discrete data group and its quantity, and number the discrete data group;

[0108] The clustering module is configured to input the driving events represented by the numbers of the discrete data groups into the LDA model, set the number of clusters for the LDA model, and cluster the driving events.

[0109] Embodiment 2

[0110] This embodiment further explains the clustering method of the technical solution with an actual driving event clustering example:

[0111] Step 1: Obtain a total of 2957 driving events

[0112] Step 2: Set 30 km / h and 60 km / h as speed thresholds, and set the speed as v.

[0113] If 0 km / h ≤ v ≤ 30 km / h, the speed category is low speed;

[0114] If 30 km / h < v ≤ 60 km / h, the speed category is medium speed;

[0115] If v > 60 km / h, the speed category is high speed.

[0116] Step 3: As Figure 2 shown, based on vehicle longitudinal dynamics, perform a coupled analysis on the speed and longitudinal acceleration of driving data points. Taking the speed v as the independent variable and the longitudinal acceleration a x as the dependent variable, output the v-a x scatter plot of driving data points; obtain the maximum longitudinal acceleration a x1 and the minimum longitudinal acceleration a x2 corresponding to each speed v, and fit to obtain the v-a x coupling shape.

[0117] This embodiment va x Upper envelope: a x,max = -0.09v + 4.11,

[0118] va x Lower envelope: a x,min =0.14v-8.00,

[0119] Set longitudinal acceleration a x The uniform velocity range is -0.05 m / s. 2 ~0.05m / s 2 ,

[0120] Obtain the upper envelope a x,max =a corresponding to velocity v on f1(v) x With 0.05m / s 2 The difference And Multiply by 50% to get: Fitting speed v and Based on the dependency relationship, the thresholds for gradual acceleration and rapid acceleration are obtained:

[0121] Obtain the lower envelope a x,min =a corresponding to the velocity v on f2(v) x With -0.05m / s 2 The difference And Multiply by 50% to get: Fitting speed v and Based on the dependency relationship, the thresholds for gradual deceleration and rapid deceleration are obtained:

[0122] Step 4: As Figure 3 As shown, based on the constraint of the tire friction circle, with longitudinal acceleration a x Let a be the independent variable and the lateral acceleration. y As the dependent variable, output 'a' for the driving data points. x -a y Scatter plot; In this embodiment, the fitted circle representing the lateral acceleration threshold is finally calculated as: a y,50％ 2 =3.25 2 -a x 2 .

[0123] Step 5: 3.6 m / s 3 and -3.6m / s 3 Let the longitudinal jerk threshold be defined, and let the longitudinal jerk be j.

[0124] If 0 m / s3 ≤j≤3.6m / s 3 If the longitudinal jerk j is , then the category is gradual acceleration;

[0125] If j > 3.6 m / s 3 If the longitudinal jerk j is , then the category is rapid acceleration;

[0126] If -3.6m / s 3 ≤j<0m / s 3 If the longitudinal jerk j is , then the category is deceleration acceleration;

[0127] If j < -3.6 m / s 3 If the longitudinal jerk j is , then the category is rapid deceleration.

[0128] The threshold values ​​for each segmentation of the driving data points in this embodiment are shown in Table 1:

[0129]

[0130]

[0131] Table 1

[0132] Step 6: Based on the classification thresholds or categories of speed, longitudinal acceleration, lateral acceleration, and longitudinal jerk in Table 1, discretize the driving data points contained in the 2957 driving events to obtain a total of 75 discrete data groups; number these 75 discrete data groups sequentially as 1, 2, 3...75.

[0133] Step 7: Input the driving events represented by these 75 discrete data groups into the LDA model in sequence, set the number of clusters for the LDA model to 5, and cluster the 2957 driving events.

[0134] This embodiment uses perplexity to characterize the performance of the LDA-improved model; the lower the perplexity, the better the model's performance. Besides using 5 clusters, this embodiment also conducted multiple experiments with different numbers of clusters, such as... Figure 4 As shown, when the number of clusters is between 2 and 5, the perplexity decreases as the number of clusters increases. When the number of clusters is greater than or equal to 5, the perplexity approaches saturation and fluctuates around 5. For this technical solution, a number of clusters of 5 is the optimal number. In summary, compared to the perplexity of around 20.7 in the classic model, the improved LDA model in this embodiment has better clustering performance.

[0135] The above examples are only used to illustrate the technical solutions of the present application, but not to limit the present application; based on the examples of the present application, all other examples obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application. Although the present application is described in detail with reference to the foregoing examples, those of ordinary skill in the art should understand that they can modify the technical solutions recorded in the foregoing examples, or make equivalent replacements for some technical features; 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 event clustering method based on an improved LDA model, characterized in that, The method comprises the following steps: obtaining a plurality of driving events, each of the driving events comprising a plurality of driving data points, each of the driving data points comprising a speed, a longitudinal acceleration and a lateral acceleration, and calculating a longitudinal jerk of the driving data point according to the longitudinal acceleration; dividing the speed into three categories of low speed, medium speed and high speed according to a preset speed threshold; performing coupling analysis on the speed and the longitudinal acceleration of the driving data point based on vehicle longitudinal dynamics, obtaining a threshold of the longitudinal acceleration under different speeds, and dividing the longitudinal acceleration into five categories of slow acceleration, rapid acceleration, constant speed, slow deceleration and rapid deceleration according to the threshold; performing coupling analysis on the longitudinal acceleration and the lateral acceleration of the driving data point based on the limitation of the tire friction circle, obtaining a threshold of the lateral acceleration under different longitudinal accelerations, and dividing the lateral acceleration into two categories of low side shift and high side shift according to the threshold; Longitudinal acceleration as a function of the lateral acceleration as an independent variable, lateral acceleration as a dependent variable, output driving data points - scatter plot; Get the point on the scatter plot that is farthest from the origin (0, 0). , ), with the origin of the scatter plot as the center, and ( , The distance between the origin and the radius is... ,get - Fitting the circle for the coupled shape: Let the fitted circle be set as - The envelope; with the origin of the scatter plot as the center of the circle, with the origin of the scatter plot as the center of the circle, , If then the lateral acceleration is of the category low sideslip; If then the lateral acceleration of the class high side shift; dividing the longitudinal jerk into four categories of slow jerk acceleration, rapid jerk acceleration, slow jerk deceleration and rapid jerk deceleration according to a preset longitudinal jerk threshold; performing discretization processing on the driving data points of the driving events based on the categories of the speed, the categories of the longitudinal acceleration, the categories of the lateral acceleration and the categories of the longitudinal jerk, to obtain a discrete data set; inputting the driving events represented by the discrete data set into an LDA model, setting a cluster number for the LDA model, and clustering the driving events.

2. The clustering method of claim 1, wherein, obtaining original data points, and performing smoothing processing on the original data points according to a moving average method to obtain driving data points; obtaining a longitudinal jerk of the driving data point, and performing smoothing processing on the longitudinal jerk according to a moving average method.

3. The clustering method of claim 1, wherein, Setting 30 km / h and 60 km / h as speed threshold, setting speed as , If then the speed category is low speed; If then the speed category is medium speed; If then the speed category is high speed.

4. The clustering method of claim 1, wherein, With speed as the independent variable, longitudinal acceleration as the dependent variable, outputting a driving data point - scatter plot; acquiring each speed corresponding maximum longitudinal acceleration and minimum longitudinal acceleration , fitting to obtain - coupling shape; based on - the variation law of the coupling shape, determine - the positive and negative of the slope of the upper envelope line of the scatter plot and the positive and negative of the slope of the lower envelope line. According to the positive and negative of the slope of the upper envelope line, the point located at - Coupling two peak points of the upper half region of the shape, based on the coordinates of the two peak points, the following is calculated - Upper envelope line: ; According to the positive and negative of the slope of the lower envelope line, the point located at - Coupling two peak points of the lower half region of the shape, based on the coordinates of the two peak points, the following is calculated - Lower envelope line: .

5. The clustering method of claim 4, wherein, Setting longitudinal acceleration of the uniform interval is -0.05 ~0.05 , If then the longitudinal acceleration is of the category slow acceleration or fast acceleration; If then the longitudinal acceleration is of the category of gentle deceleration or sharp deceleration; upper envelope upper velocity corresponding with the difference and multiplying by 50% gives: ; fitted velocity dependence of with the threshold between gradual and abrupt acceleration: ; If then the longitudinal acceleration is of the category slow acceleration; If then the longitudinal acceleration is of the category of abrupt acceleration; Obtain the lower envelope Speed corresponding and The difference and will Multiply by 50% to get: Fitting speed and Based on the dependency relationship, the thresholds for gradual acceleration and rapid acceleration are obtained: ; If then the longitudinal acceleration is of the category of deceleration. If then the longitudinal acceleration is of the category of abrupt deceleration.

6. The clustering method of claim 1, wherein, Taking the derivative of the longitudinal acceleration, obtaining the longitudinal jerk based on the derivative, setting 3.6 and -3.6 as the longitudinal jerk threshold, setting the longitudinal jerk as , If then the longitudinal jerk is of the category slow jerk; If then the longitudinal jerk is of the category of sharp jerk; If then the longitudinal jerk is of the category decelerating acceleration; If then the longitudinal jerk is of the class of hard deceleration.

7. The clustering method of claim 1, wherein, performing discretization processing on the driving data points based on the categories of the speed, the categories of the longitudinal acceleration, the categories of the lateral acceleration and the categories of the longitudinal jerk, to obtain a discrete data set; inputting the driving events represented by the discrete data set into an LDA model, setting a cluster number for the LDA model, and clustering the driving events.

8. A driving event clustering system based on an improved LDA model, characterized in that, comprise: a driving event obtaining module configured to obtain a plurality of driving events, each of the driving events comprising a plurality of driving data points, each of the driving data points comprising a speed, a longitudinal acceleration and a lateral acceleration; a longitudinal jerk output module configured to calculate a longitudinal jerk of the driving data point according to the longitudinal acceleration; a speed category dividing module configured to divide the speed into three categories of low speed, medium speed and high speed according to a preset speed threshold; a longitudinal acceleration category dividing module configured to perform coupling analysis on the speed and the longitudinal acceleration of the driving data point based on vehicle longitudinal dynamics, obtain a threshold of the longitudinal acceleration under different speeds, and divide the longitudinal acceleration into five categories of slow acceleration, rapid acceleration, constant speed, slow deceleration and rapid deceleration according to the threshold; a lateral acceleration category dividing module configured to perform coupling analysis on the longitudinal acceleration and the lateral acceleration of the driving data point based on the limitation of the tire friction circle, obtain a threshold of the lateral acceleration under different longitudinal accelerations, and divide the lateral acceleration into two categories of low side shift and high side shift according to the threshold; The longitudinal jerk category division module is configured to divide the longitudinal jerk into four categories of slow jerk acceleration, rapid jerk acceleration, slow jerk deceleration and rapid jerk deceleration according to a preset longitudinal jerk threshold value; The driving data point discretization module is configured to discretize driving data points of a driving event based on the categories of speed, longitudinal acceleration, lateral acceleration and longitudinal jerk, to obtain a discrete data group; Longitudinal acceleration based on tire friction circle limits Lateral acceleration as independent variable Longitudinal acceleration as dependent variable - Scatter plot Obtain a point farthest from the origin (0, 0) on the scatter plot , ), and obtain a fitting circle of the coupling shape with the origin of the scatter plot as the center and the distance between (x, y) and the origin as the radius , - , Set the fitting circle as the envelope line of (x, y) - .​​ with the origin of the scatter plot as the center of the circle, with the origin of the scatter plot as the center of the circle, , If then the lateral acceleration is of the class low side shift; If then the lateral acceleration is of the class high lateral displacement; The clustering module is configured to input the driving event represented by the discrete data group into an LDA model, set a clustering number for the LDA model, and cluster the driving event.

9. The clustering system of claim 8, wherein, The driving data point discretization module includes: The discrete data group editing unit is configured to obtain the discrete data group and the number thereof, number the discrete data group, and input the driving event represented by the number of the discrete data group into the LDA model.

Images