CN109367541B

CN109367541B - Intelligent vehicle-like person lane change decision-making method based on driver behavior characteristics

Info

Publication number: CN109367541B
Application number: CN201811195634.1A
Authority: CN
Inventors: 郭洪艳; 梁德宝; 刘惠; 陈虹; 刘风; 刘俊; 戴启坤
Original assignee: Jilin University
Current assignee: Jilin University
Priority date: 2018-10-15
Filing date: 2018-10-15
Publication date: 2020-12-25
Anticipated expiration: 2038-10-15
Also published as: CN109367541A

Abstract

The invention discloses an intelligent vehicle type human lane change decision method based on driver behavior characteristics, which comprises the steps of extracting parameters representing the steering characteristics of a driver by establishing a driver steering characteristic experiment, then establishing a driver characteristic recognizer by utilizing a K-means clustering and BP neural network method, and carrying out human-like lane change decision according to the recognized driver characteristics of the driver characteristic recognizer, the self and surrounding vehicle states recognized by a sensor and environmental information, so that the lane change behavior of an intelligent vehicle has human-like driving characteristics.

Description

Intelligent vehicle-like person lane change decision-making method based on driver behavior characteristics

Technical Field

The invention relates to an intelligent vehicle lane change decision method, in particular to an intelligent vehicle human lane change decision method based on the steering behavior characteristics of a driver.

Background

The amount of automobile keeping is continuously increasing worldwide, which also brings about a lot of traffic safety problems. In china for example, the total number of traffic accidents is about 470 ten thousand per year, and thus the number of deaths is about 21%. The traffic accident statistical result in 2008 shows that the traffic accident caused by wrong behavior of the driver in China is 230727, which accounts for 87% of the total number of the traffic accidents in 2008, so that 61065 people die and 265889 people are injured, which respectively account for 83.1% and 87.2% of the total number of the dead people and the total number of the injured people. Obviously, in the typical system of "man-car-road", man as the decision-making subject of the closed-loop system has a great influence on cars and roads and other things involved therein, and the influence is mainly caused by man-made reasons in most traffic accidents.

To ameliorate this problem, it is necessary to dilute the human role in the closed loop "human-vehicle-road" system, or even to change the closed loop system directly to a "vehicle-road" closed loop system. This requires that the vehicle itself be somewhat intelligent, be easier to drive, and even be able to achieve autonomous driving without human manipulation. To realize this function, research on the technology of intelligent vehicles needs to be conducted.

In the past decades, the field of smart cars has been revolutionized. The rapid alternation of hardware devices such as laser radar, millimeter wave radar, a camera and a processor enables the development speed of advanced driving auxiliary systems such as an adaptive cruise control system, a lane keeping auxiliary system and a traffic jam auxiliary system to be increased, and the development speed of intelligent vehicles is accelerated.

The intelligent vehicle needs to be capable of replacing the brain of a person to autonomously make a reasonable behavior decision according to the surrounding dynamic driving environment, and the corresponding decision-making behavior of the limbs of the person is completed through an execution mechanism. Therefore, the accurate decision-making behavior of the intelligent vehicle is a key core technology, and the decision-making behavior considering the characteristics of the driver can be more specific to the driving comfort of the passengers, and can be called as the brain of the whole system. Therefore, the human-like lane change decision is a key technology for guaranteeing the safe driving of the intelligent vehicle on the road, and the research on the human-like lane change decision problem of the intelligent vehicle is of great significance.

Disclosure of Invention

In order to solve the problems in the prior art, the invention provides an intelligent vehicle type human lane change decision method based on driver behavior characteristics, which comprises the steps of extracting parameters representing the steering characteristics of a driver by establishing a driver steering characteristic experiment, then establishing a driver characteristic recognizer by using a K-means clustering and BP neural network method, and carrying out human-like lane change decision according to the recognized driver characteristics of the driver characteristic recognizer, the states of the vehicle and the surrounding vehicles recognized by a sensor and environmental information, so that the lane change behavior of an intelligent vehicle has human-like driving characteristics.

The invention is realized by the following technical scheme:

an intelligent vehicle-type person lane change decision-making method based on driver behavior characteristics comprises the following steps:

step one, establishing a steering characteristic experiment of a driver:

establishing a simulated traffic scene in Carsim; searching a plurality of drivers as experimenters, and dividing the experimenters into a training group and a testing group;

step two, establishing a driver characteristic classifier by using a K-means clustering method:

taking the experimental data of each experimenter in the training set as a clustering sample, selecting two clustering centers, and dividing the experimental data into two types, namely an aggressive type and a conservative type, by a K-means clustering method;

step three, establishing a driver characteristic recognizer by using a BP neural network method:

taking the experimental data which are subjected to the classification labeling in the step two as a training set of the BP neural network, and training the BP neural network; testing the BP neural network by using the experimental data in the test set aiming at the trained BP neural network; the BP neural network driver characteristic recognizer passing the test is used for recognizing the driving type of the driver on line;

step four, establishing a lane change intention recognizer for recognizing the driving type of the driver;

and step five, establishing a lane change clearance selector, and judging and selecting a lane change space according to different driving types.

The intelligent vehicle human lane change decision method based on the behavior characteristics of the driver comprises the following steps of:

steering wheel rotation angles and vehicle speeds v in a plurality of turning points in an experimental road are respectively extracted_tTo obtain multiple sets of experimental data; the rotating speed v of the steering wheel is obtained from each group of dataSteering wheel angle standard deviation sigmaAnd three characteristic values of the average vehicle speed v during steering; then, three groups of data corresponding to the three characteristic values are subjected to normalization processing.

The intelligent vehicle-type person lane change decision method based on the behavior characteristics of the driver comprises the following steps of establishing a lane change intention recognizer, and recognizing the driving type of the driver, wherein the lane change intention recognizer comprises the following steps:

setting different expected vehicle speeds v for different driving types in the experimenter according to the driving type identification result of the driver in the step three_xdesAnd the weight coefficients alpha, omega₁Calculating the average speed performance U by combining the three variable parameters_lvAverage time efficiency U_ltgAnd road length efficiency U_ldComposed road total efficiency U_lAnd comparing the total performance of each lane if the total performance U of a side lane₁And if the lane is more than the total efficiency of the lane, the current vehicle is determined to have the expectation of lane change of the side lane.

The intelligent vehicle human lane change decision method based on the behavior characteristics of the driver has the total road efficiency U_lRepresented by the following formula:

wherein, U_lThe total energy efficiency of the road is s;

w₁、w₂and w₃The weight coefficients of the three lane change indexes are respectively;

N_lvregular factor for average speed performance:

N_ltgregularization factor for average time performance: n is a radical of_ltg＝αtg_des

N_ldRegularization factor for road length efficiency:

the intelligent vehicle-type human lane change decision method based on the behavior characteristics of the driver comprises the following steps of establishing a lane change gap selector, and judging and selecting lane change spaces according to different driving types, wherein the lane change gap selector comprises the following steps:

when the signal of expecting lane change is obtained in the fourth step, the driving type identified in the third step is combined, and the longitudinal vehicle distance A is calculated_dAnd the distance d between the front and rear vehicles and the vehicle_f,d_rSelecting proper lane-changing space when A_d＞A_LAnd d_f,d_r＞d_sWhen the temperature of the water is higher than the set temperature,

i.e. there is a variable lane space; wherein A is_LTo allow a longitudinal safety distance for lane change, for different driving types A_LThe values of (A) are different; d_sTo allow a safe distance between two vehicles for lane change, for different driving types d_sThe values of (A) are different; for the available lane-changing spaces, different lane-changing spaces are selected according to different driving types.

Due to the adoption of the technical scheme, the invention has the beneficial effects that:

1. different driving parameters (desired speed v) are selected for different driver characteristics_xdesCoefficient of weight α, ω₁Longitudinal safety distance A_LAnd a safe distance d between two vehicles_s) The intelligent vehicle can be more humanized in the lane changing process on the expressway, so that the riding comfort can be further improved;

2. the invention provides a new lane change clearance selector which forms an intelligent vehicle decision link together with a lane change intention recognizer, so that the lane change decision process of the intelligent vehicle is safer and more humanized.

Drawings

FIG. 1 is a Road Course path diagram;

FIG. 2 is a partial data diagram of a driver recording personnel records;

FIG. 3 is an experimental path;

FIG. 4 is a classification of driver steering characteristics after K-means clustering;

FIG. 5 is a diagram of a neural network architecture;

FIG. 6 is a classification result of a trained neural network of a validation set;

FIG. 7 is a simulated highway traffic scene;

FIG. 8 is a lane change gap selection strategy;

FIG. 9 is a schematic diagram of lane change safety clearance calculation;

FIG. 10 is a schematic view of a lane change safety gap for an aggressive driver;

fig. 11 is a schematic diagram of a conservative lane change safety gap for a driver.

Detailed Description

The technical solution proposed by the present invention will be further explained and explained with reference to the accompanying drawings.

The invention provides an intelligent vehicle-like person lane change decision method based on driver behavior characteristics, which comprises the following steps:

1. establishing a driver steering characteristic experiment:

1.1 working condition design

A simulated traffic scene is established in Carsim, and comprises lane shapes and size designs, wherein the size designs mainly comprise road widths and lane widths, and surrounding environment designs such as houses, trees and the like.

The driving simulator used in the experiment is established based on the simulation software of Carsim, so the experimental scene of the experiment is also established in Carsim. Considering that more behaviors are turning behaviors in the lane changing process, a Road Course which is provided by Carsim and contains a plurality of different turning scenes is selected in the experiment, and the path of the Road Course is shown in FIG. 1. Considering that the expressway is our landing scene, in the lane setting, the lane is designed to be a single two-lane road surface, and the road width is set to be 8m, wherein the width of the single lane is 4 m. In order to make the experimental scene more real and reliable, lawns, trees, houses, other vehicles and the like are added into the experimental scene, and the reference objects can provide speed references for the driver.

1.2 search for experimenters

A plurality of drivers with drivers' licenses, driving ages, ages and male and female proportions which are uniformly distributed are searched as experimenters of the experiment, and the selected experimenters are divided into two groups, namely a training group and a testing group.

The experimental design searches 21 drivers, wherein 17 drivers are training set samples and are used as training sets of a neural network system, and for the 17 drivers, adults with motor vehicle drivers' licenses are selected as experimenters, the driving ages are distributed between 0 and 6 years, the ages are distributed between 19 and 38 years, and the male-female ratio is 2: 1. Training set laboratory samples are shown in table 1.

TABLE 1 training set laboratory personnel samples

The remaining 4 drivers were selected as test set samples to verify the accuracy of the neural network system. The four drivers also all have a vehicle driver license, and the sample of the test set experimenters is shown in table 2.

TABLE 2 test set laboratory personnel samples

1.3 Experimental procedures and data acquisition

Experimenters sequentially carry out experiments on the driving simulator, and transmitted experimental data are stored and recorded.

First, before the experiment is formally started, experimenters are required to fill in an information acquisition table so as to know basic information and daily driving habits of drivers. And then, the experimenter is told to firstly test driving for one circle, the driver feels and adapts to the driving simulator and the traffic environment in the process, the experimenter communicates with the experimenter after testing driving for one circle, the state of the experimenter is ensured, a second circle (formal experiment) is prepared to be started, and the recording personnel collects driving data. Part of the data recorded by the recording personnel is shown in figure 2.

1.4, data processing

Processing the obtained experimental data, selecting nine turning points in the experimental road, and respectively extracting steering wheel turning angles () and () of the nine turning pointsVehicle speed (v)_t) So as to obtain nine groups of data of each driver, further processing the data, and obtaining 3 characteristic values in each group of data, wherein the 3 characteristic values are respectively the rotating speed (v) of the steering wheel) Steering wheel angle standard deviation (σ)) And the average vehicle speed (v) at the time of steering. Then, three groups of data corresponding to the three characteristic values are subjected to normalization processing.

The data obtained by the experiment are processed to extract the parameter steering wheel rotating speed (v) which can represent the steering characteristic of the driver) Steering wheel angle standard deviation (σ)) And the average vehicle speed (v) at the time of steering.

Taking an experimenter as an example, firstly extracting three kinds of data, namely time (t), steering wheel angle () and vehicle speed (v), from all data recorded about the experimenter; then, according to the same interval, intercepting 9 sections of the three data, as shown in fig. 3, wherein the bold curve section in the figure is 9 effective turning road sections selected from the experimental path as 9 groups of sample points of each driver; finally, the following three data are obtained for each group of sample points according to the formulas (1) to (3), wherein the three data are respectively the rotating speed (v) of the steering wheel) Steering wheel angle standard deviation (σ)) And the average vehicle speed (v) at the time of steering. The data of each experimenter were processed as described above. The training set obtains 153 groups of training samples, and the test set obtains 36 groups of test samples.

Wherein v isIs the steering wheel speed, deg/s;_maxin a steering process (indicating a section of the steering process of the pay-off reel from zero to the maximum value to zero, as shown by bold in FIG. 3Curve shown) maximum value of the pay-off reel angle; t is t_maxFor the first time from the start of steering (steering wheel angle of 0) to the maximum steering wheel angle (steering wheel angle of 0)_max) The time taken; sigmaIs the steering wheel angle standard deviation, °; n is the number of data; i is a cyclic variable, i belongs to (1, n);_ifor each steering wheel angle in the n-capacity data set, i ∈ (1, n), °;

is the average value of the steering wheel angle, in the data set of volume n; v is the average vehicle speed in the primary steering process, m/s; v. of_iFor each vehicle speed in the data set with the capacity n, i belongs to (1, n), m/s;

2. method for establishing driver characteristic classifier by using K-means clustering

Taking the experimental data of each driver in the training set obtained in the step 1.4) as a clustering sample, selecting two clustering centers, and dividing the experimental data into two types, namely an aggressive type (idx is 1) and a conservative type (idx is 2) by a K-means clustering method.

For 153 groups of data of 17 experimenters after data processing, the clustering input is u ═ v,σ,v]In order to satisfy the classification of the experimenters (aggressive type and conservative type), two clustering centers are selected, which respectively represent aggressive drivers (idx is 1) and conservative drivers (idx is 2), so that the clustering output is y is idx. As shown in formula (4), the Euclidean distance formula is selected as the measure for measuring the similarity of steering characteristics of the characteristic driver, namely the cluster state variable, as shown in formula (5), J is selected as the optimization target,

wherein A and B respectively represent two array matrixes; n, mDividing into the row number of A and B matrixes; i is a cyclic variable, i belongs to (1, n); j is a cyclic variable, j belongs to (1, m); j is an optimization target;₁、₂are weight coefficients respectively; x (i) is a matrix sequence comprising 153 sets of data; c₁，C₂Two cluster center matrix sequences are respectively.

The clustered experimenter types are shown in table 3, because each experimenter has nine sample points, when the steering behavior characteristic type of one experimenter is identified, the type of the driver is identified by calculating the proportion of the nine sample points of the driver in the conservative type and the aggressive type when the proportion is more than 50%, for example, the driver 1 is identified, and the proportion of 5 sample points in the conservative type is 55.6%, so the driver 1 is identified as the conservative type. Clustering results as shown in fig. 4, the cross points represent aggressive data samples, the filled dots represent conservative data samples, and the two filled pentagons represent the clustering centers of the two classes. The cluster centers are [0.1728,0.1729,0.6311] and [0.1030,0.1389,0.3306], respectively.

TABLE 3 type of steering behavior characteristics of the experimenters

3. Method for establishing driver characteristic recognizer by using BP neural network

Taking the experimental data which are subjected to the classification labeling in the step two as a training set of the BP neural network, and training the BP neural network; aiming at the trained BP neural network, testing the BP neural network by using experimental data in a test set as a test set; and for the BP neural network driver characteristic recognizer passing the test, the BP neural network driver characteristic recognizer is used for recognizing the driving type of the driver on line.

Considering the possible discontinuous function, selecting and designing a BP neural network containing two hidden layers, considering three driver steering characteristic identification parameters, and selecting the number of input layer nodes, n_iThe number of nodes of the output layer is n as 3_oCalculating to obtain the node number according to the formula (6) as 1The number range is that the number of the nodes of the first hidden layer is 5 and the number of the nodes of the second hidden layer is 3 through debugging selection, and the structure diagram of the neural network is shown in fig. 5. 153 groups of data classified by the classifier in 2 (containing v),σV, idx) as a training set of the BP neural network, i.e., train _ set ═ v,σ,v,idx]And optimally training the BP neural network according to a formula (7) as an optimization target. Trained output function h_θ(. 8) as shown in equation (9), the function of each layer as shown in equation (9), and the weight coefficient of each layer of the trained neural network

The deviation of each layer of the trained neural network is shown in the formulas (13-15) as shown in the formulas (10-12).

h_θ(u)＝g(θ⁽³⁾·g(θ⁽²⁾·g(θ⁽¹⁾·u+b⁽²⁾)+b⁽³⁾)+b⁽⁴⁾) (8)

θ⁽³⁾＝[-0.7857 5.0422 0.8290] (12)

b⁽²⁾＝[-2.4343 0.2468 -0.2453 2.5962 -1.0204]^T (13)

b⁽³⁾＝[-2.0462 -0.0586 -1.8793]^T (14)

b⁽⁴⁾＝0.7464 (15)

Wherein m is the number of nodes of the hidden layer 1 and the hidden layer 2; m is_iNumber of nodes of input layer; m is_oNumber of output layer nodes; a is coefficient, a is equal to [1,10 ]](ii) a J (theta, b) is an optimization target of the neural network; n is the number of samples; i is a cyclic variable, i belongs to (1, n); y is⁽ⁱ⁾To output the sequence, y⁽ⁱ⁾＝idx；h_θ(. h) is an output function,. is an arbitrary value; u. of⁽ⁱ⁾For input samples u⁽ⁱ⁾＝[v,σ,v](ii) a λ is a weighting factor, λ ═ 1; l is the second layer, L is equal to [1, L-1 ]](ii) a L is the total number of layers of the neural network, and L is 4; n is the number of samples; k is a cyclic variable, k is an element (1, S)_l)；S_lIs a dimensionless book, S when l is 1_lWhen l is 2, S is 3_lWhen l is 3, S is 5_l3; j is a cyclic variable, j is an element (1, S)_l+1)；S_l+1Is a dimensionless book, S when l is 1_l+1When l is 2, S is 5_l+1When l is 3, S is_l+1＝1；

For the weight coefficient of the neural network, as shown in a formula (10-12), a method is found, namely finding a corresponding matrix through i at the upper right corner, and then finding a corresponding element according to jk at the lower right corner, wherein j is an abscissa, and k is an ordinate; g (-) is a function of each layer,. is an arbitrary value; u is input matrix u ═ v,σ,v]；θ⁽¹⁾、θ⁽²⁾And theta⁽³⁾Respectively are the weight from an input layer to a hidden layer 1, the weight from the hidden layer 1 to a hidden layer 2 and the weight from the hidden layer 2 to an output layer; b⁽²⁾、b⁽³⁾And b⁽⁴⁾The hidden layer 1, the hidden layer 2 and the output layer are respectively offset.

Then use 36 sets of data in the verification set test _ set [ v ],σ,v]The trained BP neural network is tested, and the basic information of the verification set personnel is shown in Table 4. The result of the classification of the verification set is shown in FIG. 6, and the driver characteristic recognizer recognizesThe driving types of the four drivers are identified as shown in table 5, and it is found from the test results that the driving characteristics of the four drivers can be correctly identified. Illustrating that the driver characteristic recognizer we have designed is available.

Table 4 verification set personnel basic information table

Numbering	Sex	Age of driver	Age (age)	Type (B)
					1	For male	1	25	Radical advancement
2	For male	0.5	29	Conservative
					3	For male	5	33	Radical advancement
4	For male	2	27	Conservative

TABLE 5 type of steering behavior characteristics for verification set personnel

4. Channel-changing will recognizer

Setting different expected vehicle speeds v for different driving types of experimenters_xdesAnd the weight coefficients alpha, omega₁The average velocity performance U is calculated from three main parameters by combining the three variable parameters_lvAverage time efficiency U_ltgAnd road length efficiency U_ldComposed road total efficiency U_lAnd comparing the total performance of each lane if the total performance U of a side lane₁Greater than the total efficiency U of the lane₀Then the current vehicle is determined to have the desire for lane change of the side lane.

In the step, the intelligent vehicle is mainly used for determining whether to have a lane change intention according to the information of the intelligent vehicle, surrounding vehicles and environment like a person. It is assumed that the sensor can sense the vehicle information and the environment information around the sensor, and the information is processed and transmitted to the lane change intention recognizer for use. Considering that lane change will is mainly based on different points of two adjacent lanes, the different points in the two roads can be mainly summarized as a following three points, firstly, the vehicle speed is the vehicle speed, and the average vehicle speeds of the two adjacent lanes are different; then, the time interval between the vehicles is set, and the average vehicle distance between two adjacent lanes is also different; and finally whether the road has an end. The three are different points of the two lanes and are three main factors to be considered in the lane changing process.

4.1 average velocity efficiency

Considering that the speed difference between adjacent lanes is a lane change index, we refer to the concept of average speed performance, as shown in equation (16).

Wherein, U_lvAverage velocity performance, s; d_maxFor the maximum distance to be traveled, d is a constant value_max＝6000m；v_xdesThe desired vehicle speed for the driver, m/s; gamma is a constant which makes the denominator of the equation not zero, and gamma is 5 m/s; v. of_lμIs the average speed of the road, m/s.

As can be seen from equation (16), the average vehicle speed v on the road_lμCloser to the driver's desired vehicle speed v_xdesThe higher the desire to change lanes by the driver, and the smaller the desire to change lanes.

Therefore, for drivers of different driving types, each type of driver has own expected speed, so that for drivers of different driving types, different expected vehicle speeds are selected, for aggressive drivers, high vehicle speeds are selected, and for conservative drivers, low vehicle speeds are selected relative to aggressive drivers. For the purpose of design validation, in the present invention, we select a desired vehicle speed v for an aggressive driver _xdes30m/s, while for conservative drivers we choose the desired vehicle speed v_xdes＝20m/s。

4.2 mean time efficiency

Considering that the time interval difference between adjacent lanes is a lane change index, we refer to the concept of average time performance, as shown in equation (17).

U_ltg＝min(αtg_des,tg_lμ) (17)

Wherein, U_ltgAverage time performance, s; alpha is a weight coefficient of a time interval between workshops; tg_desA desired headway; tg_lμAnd the average time interval of the vehicles on the road is obtained.

As can be seen from equation (17), the average headway of the road can be increased, and when the average headway reaches the desired headway, the maximum value of the average time efficiency is reached, and the effect of the lane change which is most desired is reached.

Therefore, for drivers of different driving types, each type of driver has own inter-vehicle time distance, different inter-vehicle time distance weight coefficients, namely different alpha, are selected for the drivers of different driving types, a relatively small inter-vehicle time distance coefficient can be selected for aggressive drivers, and a large inter-vehicle time distance coefficient relative to aggressive drivers can be selected for conservative drivers. The design effect is verified, and in the invention, for an aggressive driver, the expected workshop time distance coefficient selected by the driver is 1; for conservative drivers, the desired headway factor we have chosen is 2.

4.3 road length efficiency

Considering whether the end of a road exists adjacent to a road as a lane change index, we refer to the concept of road length efficiency, as shown in equation (18).

Wherein, U_ldAverage time performance, s; d_minTo account for the end of the road, the minimum distance the vehicle is allowed to travel.

As can be seen from the equation (18), the smaller the distance of the end of the road is, the smaller the road length efficiency is, and conversely, the larger the distance of the end of the road is, the larger the road length efficiency is, the better the vehicle is suitable for running, and the lane change expectation of the vehicle is also increased.

4.4 Total road efficiency

In view of the above three lane change indicators, we refer to the total energy efficiency of the road, as shown in equation (19). Total energy efficiency U of adjacent lane_lU larger than the lane_lThe vehicle has a willingness to change lanes of the adjacent lanes.

Wherein, U_lThe total energy efficiency of the road is s; w is a₁、w₂And w₃The weight coefficients of the three lane change indexes are respectively; n is a radical of_lvThe regularization factor, which is the average velocity performance, is shown in equation (20); n is a radical of_ltgThe regularization factor, which is the mean time performance, is shown in equation (21); n is a radical of_ldThe regularization factor for road length efficiency is shown in equation (22).

N_ltg＝αtg_des (21)

For equation (19), different weighting factors w are selected for different driver types, considering that speed has a large influence in both headway and desired speed for drivers of different driving types₁For aggressive driver w₁Will be relatively small, w for conservative drivers₁The value of (c) may be greater relative to aggressive drivers. In the verification of the effect of the design, in the present invention, when the average vehicle speed is equal to or less than the desired vehicle speed, i.e., v_lμ≤v_xdesFor aggressive drivers, we choose w₁Is 1; while for conservative drivers, we choose w₁Is 1.5; when the average vehicle speed is greater than the desired vehicle speed, i.e., v_lμ＞v_xdesFor aggressive drivers, we choose w₁Is 5; while for conservative drivers, we choose w₁Is 10.

As shown in FIG. 7, an experimental scenario simulating an expressway is shown, where a road is a single-way double-row lane, vehicles all run at a constant speed, E is a vehicle, and S is₁，S₂And S₃As are other vehicles around. And the adjacent lanes of conservative type and aggressive type are calculated according to the secondary traffic sceneAt different tg_lμAnd v_lμThe total performance of the roads is shown in tables 6 and 7.

TABLE 6 conservative track-changing willingness table

TABLE 7 radical lane change wish list

In tables 6 and 7, 0.59 and 0.47 at the top right corner are respectively conservative and aggressive vehicle lane effect values, the aggressive vehicle lane effect value under the same working condition is lower, then the adjacent lane effect value is seen, the two tables can be seen, the aggressive adjacent lane effect value is larger than the conservative adjacent lane effect value, and the aggressive vehicle lane change is more likely than the conservative vehicle lane change under the same condition.

5. Selector for establishing lane change gap

The lane change gap selection strategy is shown in fig. 8.

First, when it is determined that there is a sufficient lane Change space for the lane Change of the host vehicle, i.e., when Change is 1, the lane Change of the host vehicle is calculated, and then, as shown in fig. 9, the horizontal inter-vehicle distance a in the x direction is calculated according to the equations (23), (24), and (25)_dAnd the distance d from the right front point of the rear vehicle to the mass center of the vehicle during lane changing₂And the distance d from the left rear point of the front vehicle to the mass center of the vehicle during lane changing₁。

Wherein A is_dIs the horizontal x-direction car-to-car distance, m; t is t_cLane change time, s;

is S₁Vehicle speed in the x-direction, m/s;

for the start of lane change S₁The centroid position abscissa of the vehicle, m; l is the vehicle length, m; v. of_xEThe vehicle speed of E in the x direction, m/s; d₁The distance from the left rear point of the front vehicle to the mass center of the vehicle during lane changing;

is S₁Vehicle speed in the y direction, m/s;

for the start of lane change S₁The centroid position of the vehicle is the ordinate, m; w is the vehicle width, m; v. of_yEThe vehicle speed of E in the y direction, m/s; d₂The distance from the right front point of the rear vehicle to the mass center of the vehicle during lane changing;

is S₂Vehicle speed in the y direction, m/s;

for the start of lane change S₂The centroid position of the vehicle is the ordinate, m;

is S₂Vehicle speed in the x-direction, m/s;

for the start of lane change S₂The centroid position abscissa of the vehicle, m;

then, it is judged that during the lane change, A_dIs > 1, and d₁＞1，d₂If the current working condition is more than 1, the lane change space is considered to be enough, and as shown in fig. 10 and 11, the lane change space is an aggressive type available lane change space and a conservative type available lane change space under the current working condition.

And finally, when the available lane change space is calculated, different lane change spaces are selected according to different types of drivers, the lane change space with the shortest driving time at the most front position is selected for an aggressive driver, and the lane change space with the largest lane change space is selected for a conservative driver. Under the working condition, both the aggressive type and the conservative type have one lane change space, so the lane change space is directly selected. For the case of no lane change space, a deceleration instruction is given to wait for the next calculation.

Claims

1. An intelligent vehicle-like person lane change decision-making method based on driver behavior characteristics is characterized by comprising the following steps:

step one, establishing a steering characteristic experiment of a driver:

establishing a simulated traffic scene in Carsim; searching a plurality of drivers as experimenters, and dividing the experimenters into a training set and a testing set; the experimental data processing comprises the following processes:

steering wheel rotation angles and vehicle speeds v in a plurality of turning points in an experimental road are respectively extracted_tTo obtain multiple sets of experimental data; the rotating speed v of the steering wheel is obtained from each group of dataSteering wheel angle standard deviation sigmaAnd three characteristic values of the average vehicle speed v during steering; then, carrying out normalization processing on three groups of data corresponding to the three characteristic values;

step four, establishing a lane change intention recognizer, and recognizing the driving type of a driver:

setting different expected vehicle speeds v for different driving types in the experimenter according to the driving type identification result of the driver in the step three_xdesAnd the weight coefficients alpha, omega₁Calculating the average speed performance U by combining the three variable parameters_lvAverage time efficiency U_ltgAnd road length efficiency U_ldComposed road total efficiency U_lAnd comparing the total performance of each lane if the total performance U of a side lane₁If the lane is more than the total efficiency of the lane, the current vehicle is determined to have the expectation of lane change of the side lane;

the total efficiency U of the road_lRepresented by the following formula:

wherein, U_lThe total energy efficiency of the road is obtained;

N_lvregular factor for average speed performance:

N_ldRegularization factor for road length efficiency:

wherein d is_maxThe maximum driving distance of the road is obtained; v. of_xdesThe desired vehicle speed for the driver, m/s; γ is a constant that makes the denominator of the equation non-zero; tg_desA desired headway;

step five, establishing a lane change clearance selector, and judging and selecting a lane change space according to different driving types:

when the signal of expecting lane change is obtained in the fourth step, the driving type identified in the third step is combined, and the longitudinal vehicle distance A is calculated_dAnd the distance d between the front and rear vehicles and the vehicle_f,d_rSelecting proper lane-changing space when A_d＞A_LAnd d_f,d_r＞d_sThen, there is a variable lane space; wherein A is_LTo allow a longitudinal safety distance for lane change, for different driving types A_LThe values of (A) are different; d_sTo allow a safe distance between two vehicles for lane change, for different driving types d_sThe values of (A) are different; for the available lane-changing spaces, different lane-changing spaces are selected according to different driving types.