阅读说明：本技术 一种基于综合模型检测驾驶员风险等级的方法 (Method for detecting risk level of driver based on comprehensive model ) 是由 牛世峰 董兆晨 郑佳红 付锐 郭应时 袁伟 于 2019-09-26 设计创作，主要内容包括：本发明公开了一种基于综合模型检测驾驶员风险等级的方法,本发明通过采集车辆的速度数据以及驾驶员的不同的报警类型,对驾驶行为进行判断并对驾驶员风险等级进行分类,利用训练好的综合模型对驾驶员风险等级进行识别,再利用识别结果判断驾驶员风险等级。本发明主要服务于运输企业安全管理系统,当识别到驾驶员是高风险等级的驾驶员时,可采取相应的管理培训措施,以提高其安全性。本发明的实用化可以降低因交通事故导致的人员伤亡和财产损失,提高交通系统的整体安全性。(The invention discloses a method for detecting the risk level of a driver based on a comprehensive model, which judges the driving behavior and classifies the risk level of the driver by acquiring the speed data of a vehicle and different alarm types of the driver, identifies the risk level of the driver by using the trained comprehensive model, and judges the risk level of the driver by using the identification result. The invention is mainly used for a safety management system of a transportation enterprise, and when the driver is identified to be a high-risk driver, corresponding management training measures can be taken to improve the safety of the driver. The practicability of the invention can reduce the casualties and property loss caused by traffic accidents and improve the overall safety of the traffic system.)

1. A method for detecting a risk level of a driver based on a comprehensive model is characterized by comprising the following steps:

acquiring speed data of a driver to obtain a driving behavior sequence and a driver risk level;

repeatedly and iteratively training the comprehensive model, and determining each parameter of the model to obtain the comprehensive model;

step three, detecting a driving behavior sequence to be recognized by using the trained comprehensive model;

and step four, counting the detection results in a preset time period to obtain a final risk grade judgment result under the preset confidence level of the driver.

2. The method for detecting the risk level of the driver based on the comprehensive model as claimed in claim 1, wherein in the step one, the acceleration data of the vehicle is calculated according to the speed data, and the calculation method of the acceleration data is as follows:

v_tvelocity at time t, v_t-1Speed at time t-1, Δ t being the time interval, obtained from GPS data of the vehicle, a_tIs the average acceleration over time t.

3. The method for detecting the risk level of the driver based on the comprehensive model as claimed in claim 1, wherein in the step one, when the driving behavior is determined, firstly, the acceleration information of the driver is analyzed according to the speed data of the driver, then, the driver behavior is divided into a plurality of types according to the acceleration information, the alarming times of the driver are collected according to an alarming system built in the vehicle, and the risk level of the driver is clustered through K-means.

4. The method for detecting the risk level of the driver based on the comprehensive model as claimed in claim 1, wherein the comprehensive model established in the first step comprises:

the number of hidden states is N, and the set of hidden states is Q ═ Q₁，q₂，q₃Corresponding to a low risk level, a medium risk level, a high risk level, respectively;

the number of observation states M, the set of observation states is V ═ V₁，v₂，...v₅Corresponding to fast deceleration, slow deceleration, normal driving, slow acceleration and fast acceleration respectively;

hidden state sequence is i₁，i₂，...i_TObservation sequence is o₁，o₂，...o_TAnd T is the length of the driving behavior sequence;

output probability matrix B, B ═ B_j(o_k)＝P(o_t＝v_k|i_t＝q_j) I is more than or equal to 1 and less than or equal to 3, k is more than or equal to 1 and less than or equal to 5, and the hidden state at the time t is q_jThe observed value generated is v_kMatrix B reflects the relation between the hidden state and the observed value;

the probability matrix C of the transition of the observed values,representing a risk rating of q_jIs observed by v_kConversion to v_lJ ═ 1,2, 3;

initial state distribution probability of pi ═ pi_i}，1≤i≤3，

5. The method for detecting the risk level of the driver based on the comprehensive model as claimed in claim 1, wherein in the second step, the method for training the stable comprehensive model comprises the following steps: and taking the driving behavior of the driver with each risk level as the input of the comprehensive model, and solving the model parameters of the comprehensive model.

6. The method for detecting the risk level of the driver based on the comprehensive model as claimed in claim 1, wherein in the third step, the driving behavior sequence is recognized by using the trained comprehensive model, and the output result is Selecting the risk grade corresponding to the comprehensive model with the maximum probability as the recognition result of the driving behavior sequence, wherein P is_iUnder the condition of a driving behavior sequence O and a model parameter lambda, a hidden state i at the time t_tIs a risk class q_iThe probability of (d); alpha is alpha_t(i) Is the forward probability, β, of the i state at time t_t(i) Refers to the backward probability of the i state at time t.

7. The method for detecting the risk level of the driver based on the comprehensive model as claimed in claim 1, wherein the concrete method of the step four is as follows:

firstly, judging the risk level of a driver in a short time period according to the driving behavior sequence recognition result with the length of T, wherein the judgment method comprises the following steps:

wherein the content of the first and second substances,for the recognition of the risk level of the driver in a short time,respectively representing a low risk level, a medium risk level and a high risk level, P₁、P₂、P₃Respectively representing the probabilities of identifying the driving behavior sequence to be tested as a low risk level, a medium risk level and a high risk level through a comprehensive model;

and secondly, calculating the number of the three risk levels in a long time according to the judgment result of the risk level of the driver in a short time period, wherein the calculation formula is as follows:

n1 is the number of sequences identified as low risk levels over a long period of time, n2 is the number of sequences identified as medium risk levels over a long period of time, n3 is the number of sequences identified as high risk levels over a long period of time, n is the total, x_jUsed as a count;

thirdly, calculating the proportion of each risk grade according to the number of the three risk grades in a long time, wherein the calculation formula is as follows:

k₁、k₂、k₃respectively representing the recognition proportions of a low risk grade, a medium risk grade and a high risk grade in a long time period;

fourthly, expressing the uncertainty of the random variable by using the information entropy, and calculating the formula as follows:

c is 1-H and represents the confidence coefficient of the risk level of the driver, and the higher the C value is, the higher the confidence coefficient is;

fifthly, under the condition that the confidence coefficient is C, the method for judging the risk level of the driver in a long time period is as follows:

indicating the driver's long-term risk level recognition result, I₁、I₂、I₃Respectively representing a low risk level, a medium risk level and a high risk level.

Technical Field

The invention belongs to the field of traffic safety, and particularly relates to a method for detecting a risk level of a driver based on a comprehensive model.

Background

With the rapid development of national economy and the acceleration of urbanization process, the motor vehicle ownership and road traffic volume in China are rapidly increased, and the problem of traffic accidents is increasingly prominent. Research has shown that, the driver factors are the main causes of traffic accidents, and the drivers have different driving risk levels and different contributions to traffic accidents, so that the drivers with low risk levels may cause less or even avoid traffic accidents, and the drivers with higher risk levels may cause more serious traffic accidents. Therefore, it is important to intensively study a driver risk level identification method.

At present, existing achievements mainly aim at the assessment of the risk level of the driver for research, and few researches aim at the identification of the risk level of the driver, so that the requirements of traffic safety management cannot be met.

Disclosure of Invention

The invention aims to overcome the defects and provide a method for detecting the risk level of a driver based on a comprehensive model, which can identify the risk level of the driver according to different driving behaviors and provide reference for reducing the occurrence of traffic accidents.

In order to achieve the above object, the present invention comprises the steps of:

acquiring speed data of a driver to obtain a driving behavior sequence and a driver risk level;

repeatedly and iteratively training the comprehensive model, and determining each parameter of the model to obtain the comprehensive model;

step three, detecting a driving behavior sequence to be recognized by using the trained comprehensive model;

and step four, counting the detection results in a preset time period to obtain a final risk grade judgment result under the preset confidence level of the driver.

In the first step, acceleration data of the vehicle is calculated according to the speed data, and the calculation method of the acceleration data comprises the following steps:

v_tvelocity at time t, v_t-1Speed at time t-1, Δ t being the time interval, obtained from GPS data of the vehicle, a_tIs the average acceleration over time t.

In the first step, when the driving behavior is determined, firstly, the acceleration information of a driver is analyzed according to the speed data of the driver, then the driver behavior is divided into a plurality of types according to the acceleration information, the alarming times of the driver are collected according to an alarming system arranged in a vehicle, and the risk level of the driver is clustered through k-means.

In the first step, the established comprehensive model comprises the following steps:

the number of hidden states is N, and the set of hidden states is Q ═ Q₁,q₂,q₃Corresponding to a low risk level, a medium risk level, a high risk level, respectively;

the number of observation states M, the set of observation states is V ═ V₁,v₂,…v₅Corresponding to fast deceleration, slow deceleration, normal driving, slow acceleration and fast acceleration respectively;

hidden state sequence is i₁,i₂,...i_TObservation sequence is o₁,o₂,...o_TAnd T is the length of the driving behavior sequence;

output probability matrix B, B ═ B_j(o_k)＝P(o_t＝v_k|i_t＝q_j) I is more than or equal to 1 and less than or equal to 3, k is more than or equal to 1 and less than or equal to 5, and the hidden state at the time t is q_jThe observed value generated is v_kMatrix B reflects the relation between the hidden state and the observed value;

the probability matrix C of the transition of the observed values,representing a risk rating of q_jIs observed by v_kConversion to v_lJ ═ 1,2, 3;

initial state distribution probability of pi ═ pi_i}，1≤i≤3,

In the second step, the method for training the stable comprehensive model comprises the following steps: and taking the driving behavior of the driver with each risk level as the input of the comprehensive model, and solving the model parameters of the comprehensive model.

In the third step, the driving behavior sequence is identified by utilizing the trained comprehensive model, and the output result isSelecting the risk grade corresponding to the comprehensive model with the maximum probability as the recognition result of the driving behavior sequence, wherein P is_iIs a hidden state i at the time t under the condition of known observation value sequence (driving behavior sequence) O and model parameter lambda_tIs a risk class q_iThe probability of (d); alpha is alpha_t(i) Is the forward probability, beta, of the state at time t_t(i) Refers to the backward probability of the i state at time t.

The concrete method of the step four is as follows:

firstly, judging the risk level of a driver in a short time period according to the driving behavior sequence recognition result with the length of T, wherein the judgment method comprises the following steps:

wherein the content of the first and second substances,for the recognition of the risk level of the driver in a short time,respectively representing a low risk level, a medium risk level and a high risk level, P₁、P₂、P₃Respectively representing the identification of the driving behavior sequence to be tested as a low risk grade, a medium risk grade andprobability of high risk level;

and secondly, calculating the number of the three risk levels in a long time according to the judgment result of the risk level of the driver in a short time period, wherein the calculation formula is as follows:

n1 is the number of sequences identified as low risk levels over a long period of time, n2 is the number of sequences identified as medium risk levels over a long period of time, n3 is the number of sequences identified as high risk levels over a long period of time, n is the total, x_jUsed as a count;

thirdly, calculating the proportion of each risk grade according to the number of the three risk grades in a long time, wherein the calculation formula is as follows:

k₁、k₂、k₃respectively representing the recognition proportions of a low risk grade, a medium risk grade and a high risk grade in a long time period;

fourthly, expressing the uncertainty of the random variable by using the information entropy, and calculating the formula as follows:

c is 1-H and represents the confidence coefficient of the risk level of the driver, and the higher the C value is, the higher the confidence coefficient is;

fifthly, under the condition that the confidence coefficient is C, the method for judging the risk level of the driver in a long time period is as follows:

indicating the driverIdentification of the Risk level over a prolonged period of time, I₁、I₂、I₃Respectively representing a low risk level, a medium risk level and a high risk level.

Compared with the prior art, the method and the device have the advantages that the driving behavior is judged and the risk level of the driver is classified by acquiring the speed data of the vehicle and different alarm types of the driver, the risk level of the driver is identified by using the trained comprehensive model, and then the risk level of the driver is judged by using the identification result. The invention is mainly used for a safety management system of a transportation enterprise, and when the driver is identified to be a high-risk driver, corresponding management training measures can be taken to improve the safety of the driver. The practicability of the invention can reduce the casualties and property loss caused by traffic accidents and improve the overall safety of the traffic system.

Drawings

FIG. 1 is a diagram of an integrated model of the present invention.

Detailed Description

The invention is further described below with reference to the accompanying drawings.

Referring to fig. 1, the present invention first designs an acceleration index by analyzing speed data of each driver, and identifies a driving behavior sequence of the driver by using a characteristic index. And then, dividing the drivers into low-risk level drivers, medium-risk level drivers and high-risk level drivers by using the acquired alarm type of each driver and adopting a K-Means mean clustering method. The classification of the risk level of the driver is the basis of the identification of the risk level of the driver, and whether the classification is reasonable or not directly determines the success or failure of the identification algorithm. And training the comprehensive model by using partial data, and then identifying the driving behavior sequence.

The driving behaviors comprise 5 types of rapid deceleration, slow deceleration, normal driving, slow acceleration and rapid acceleration.

1. Calculating the acceleration of the vehicle;

the vehicle acceleration is an average acceleration of the vehicle in a short time, and is specifically represented by formula (1).

In the formula: v. of_tVelocity at time t, v_t-1Speed at time t-1, Δ t being the time interval, obtained from GPS data of the vehicle, a_tIs the average acceleration over time t.

2. Determining driving behaviors and a risk level of a driver;

the acceleration threshold for distinguishing different driving behaviors is not yet specified, and the acceleration determination and discrimination method calculated according to the formula (1) is as follows:

a_tand 1,2,3, 4 and 5 represent fast deceleration, slow deceleration, normal driving, slow acceleration and fast acceleration respectively for the average acceleration in the period t. The specific determination method is to take 30% of acceleration and deceleration and 70% of acceleration and deceleration as threshold values in the acceleration sequence from large to small.

Determining the risk level of the driver mainly divides the drivers into low-risk level drivers, middle-risk level drivers and high-risk level drivers by using a K-Means clustering method for different driving alarm types.

3. Detecting a driver risk level based on the comprehensive model;

firstly, a stable comprehensive model is trained, and the specific method is that the driving behaviors of drivers with all risk levels are used as the input of the comprehensive model, the model parameters of the comprehensive model are solved, wherein lambda is { B, C, pi }, B is an output matrix, C is an observed value transfer matrix, and pi is an initial state matrix.

Then, the trained model is used for identifying the driving behavior sequence, and the output result is as follows: selecting the wind corresponding to the comprehensive model with the maximum probabilityThe risk level is used as a recognition result of the driving behavior sequence.

The method is characterized in that the detection result is counted in a certain time period to obtain a final risk grade judgment result of the driver under a certain confidence coefficient, and the method comprises the following steps:

judging the risk level of the driver in a short time period according to the driving behavior sequence recognition result with the length of T, wherein the judgment method is as follows (2):

in the formula (I), the compound is shown in the specification,a recognition result indicating a risk level of the driver for a short time,respectively representing a low risk level, a medium risk level and a high risk level, P₁、P₂、P₃Respectively representing the probability of identifying the driving behavior sequence to be tested as a low risk level, a medium risk level and a high risk level through the comprehensive model.

Secondly, calculating the number of the three risk levels in a long time according to the judgment result of the risk level of the driver in a short time period, wherein the calculation formula is as follows (3):

where n1 is the number of sequences identified as low risk levels over a long period of time, n2 is the number of sequences identified as medium risk levels over a long period of time, n3 is the number of sequences identified as high risk levels over a long period of time, n is the total,is the recognition result of the risk level of the driver in a medium-short time, x_jAnd is used only as a count.

Thirdly, calculating the proportion of each risk grade according to the number of the three risk grades in a long time, wherein the calculation formula is as shown in formula (4):

in the formula, k₁、k₂、k₃Respectively representing the recognition proportions of a low risk level, a medium risk level and a high risk level in a long time period.

Fourthly, expressing the uncertainty of the random variable by using the information entropy, wherein the calculation formula is as shown in the formula (5):

let C be 1-H, which represents the confidence in the driver risk level, with higher C values giving higher confidence.

And fifthly, under the condition that the confidence is C, judging the risk level of the driver in a long time period according to the formula (6):

in the formula (I), the compound is shown in the specification,indicating the driver's long-term risk level recognition result, I₁、I₂、I₃Respectively representing a low risk level, a medium risk level and a high risk level.

Whether the identification result of the risk grade of the driver is accurate or not is possibly related to the selection of a plurality of factors, so that the applicant carries out data analysis on the division of the driving behaviors and the classification of the risk grade based on the driving actual measurement data of the driver (n is greater than 30), and researches show that when the length of the detection sequence is 8, the detection result is relatively stable, so that when the model is trained and detected, the length of the selected driving behavior sequence is 8. In the identification considering the risk level of the driver over a long period of time, the present invention selects a period of time as long as 24 hours. The final recognition result is used as a reference according to the given confidence C value.

The method provided by the invention has the advantages that the GPS data of the vehicle and different types of alarms of the driver are collected in real time, the comprehensive model is trained and the driving behavior sequence is detected by utilizing the model through the division of the driving behavior and the classification of the risk grade, the inaccuracy and instability of the detection result based on a short period of time are overcome, the risk grade of the driver is considered through the detection of a long period of time, and the corresponding confidence coefficient is given to the final recognition result. The invention is mainly used for a vehicle safety system and a transportation enterprise safety management system, the hidden danger degrees brought to traffic safety by drivers with different risk levels are different, and the transportation enterprise can reasonably carry out batch, classification and education with different degrees on the management level of the drivers, thereby reducing the accident frequency of operating vehicles, bringing convenience for the management of the transportation enterprise and improving the operation benefit of the transportation enterprise. In addition, the invention can give reasonable advice to the driving state of the driver when the driving state fluctuates for a long time, thereby improving the safety of the driver. The practicability of the invention can reduce the casualties and property loss caused by traffic accidents and improve the overall safety of the traffic system.