CN-121980784-A - Dangerous driver longitudinal model construction method based on Lewy random characteristics

CN121980784ACN 121980784 ACN121980784 ACN 121980784ACN-121980784-A

Abstract

The invention discloses a dangerous driver longitudinal model construction method based on Lewy random characteristics. Firstly, collecting mass driving data of vehicles in an actual road traffic environment, deducing driving intensity by using a driving condition Markov state equation, adopting Lev distribution to fit the driving intensity to generate a driving intensity random sample, secondly, calculating a driving intensity boundary of current acceleration, finding an index of the driving intensity random sample on the boundary, reserving a long tail value of the index to ensure a rapid transfer path of a step length of the driving intensity, randomly selecting the long tail value as the driving intensity at the current moment, acquiring acceleration at the next moment, restraining the acceleration and calculating the speed at the next moment, and finally, updating the current speed and the current acceleration to judge whether the maximum duration is reached or collision occurs or not until a dangerous longitudinal driving behavior sequence is output. The method can quickly generate the random dangerous driver longitudinal behaviors and provide references for the simulation test of the acceleration automatic driving longitudinal auxiliary driving system.

Inventors

ZHANG MAN
AN JIANXIN
LI SONGEN
PEI ZHENLONG
ZHANG HAORAN
DOU HONGJIAN

Assignees

西安工业大学

Dates

Publication Date: 20260505
Application Date: 20260116

Claims (7)

1. A dangerous driver longitudinal model construction method based on the random Lewy characteristics is characterized by comprising the following steps: Step S1, generating a Lewy random sample meeting the real driving intensity distribution, wherein the specific steps comprise: S11, collecting mass driving data of vehicles in an actual road traffic environment, and deducing a driving intensity formula by using a driving condition Markov state equation to obtain the mass driving intensity driving data; Step S12, fitting the real driving intensity by adopting the Lewy distribution; step S13, random sampling is carried out by using the Lewy distribution, a random sample conforming to the real driving intensity distribution is generated, and step S2 is executed; step S2, randomly generating the longitudinal behaviors of the dangerous driver at the next moment, wherein the specific steps comprise: step S21, setting initial conditions; S22, calculating a driving intensity boundary of the current acceleration by using a driving condition Markov dynamics theory formula, finding an index of a driving intensity Lev random sample on the boundary, reserving a long tail value of the index, randomly selecting driving intensity corresponding to the long tail value as the current driving intensity, and acquiring the acceleration at the next moment; S23, restraining acceleration at the next moment by using the current speed boundary, calculating the speed at the next moment by using a longitudinal running equation, and executing the step S3; Step S3, outputting a longitudinal behavior sequence of a dangerous driver, wherein the specific steps comprise: Step S31, updating the current speed and the current acceleration; And S32, judging whether the current time reaches the maximum duration or whether collision occurs, if so, outputting a dangerous longitudinal driving behavior sequence, otherwise, returning to the step S2, and continuously generating a dangerous longitudinal behavior at the next moment.
2. The method for constructing a longitudinal model of a driver at risk based on random Lev characteristics of claim 1, wherein in step S11, the massive driving data of the vehicle specifically includes driving state data such as speed and acceleration, and the driving condition Markov state equation is represented by formula (1) (1) The derived driving strength formula is (2) (2)。
3. The method for constructing a longitudinal model of a dangerous driver based on the random Lewy character of claim 1, wherein in the step S12, the Lewy distribution fitting method is based on a percentile-ECF iterative estimation method, and the Lewy distribution parameters are estimated.
4. The method for constructing a longitudinal model of a dangerous driver based on the random Lewy features of claim 1, wherein in the step S13, the Lewy distribution sampling method is based on a CMS-Weron algorithm to generate a random sample of driving intensity with random Lewy features N is the number of samples.
5. The method for constructing a longitudinal model of a dangerous driver based on the random Lewy character of claim 1, wherein in the step S22, the step of finally obtaining the acceleration at the next moment comprises the following specific steps: 1) Driving intensity boundary calculation of current acceleration, defining maximum acceleration And maximum deceleration If (if) Driving intensity boundary Otherwise driving intensity boundary ; 2) Searching the index of the driving strength Lev random sample at the boundary and the corresponding driving strength value by using a formula (3), and determining the corresponding driving strength value by using a formula (4); (3) (4) 3) The long tail of the driving intensity value is reserved, a quantity deletion proportion P is defined, a deleted cut-off value T is determined, and the final driving intensity value is reserved according to the following formula: (5) (6) (7) (8) (9) (10) (11) 4) The acceleration at the next moment is randomly generated, at the final driving intensity value Randomly, and generating acceleration at the next moment by using the formula (1).
6. The method for constructing a longitudinal model of a dangerous driver based on the random Lewy character of claim 1, wherein in the step S23, the constraint mode specifically comprises: Defining a minimum maximum speed, randomly selecting acceleration at the next moment in a range from 0 to the maximum acceleration when the current speed is smaller than the minimum speed, and randomly selecting acceleration at the next moment in a range from the maximum deceleration to 0 when the current speed is larger than the maximum speed; The next time speed calculation uses the formula (12) specifically: (12)。
7. The method for constructing a longitudinal model of a driver at risk based on the random Lewy character of claim 1, wherein in the step S31, updating the current speed and the current acceleration comprises: Assigning the acceleration and speed variable at the next moment to the current acceleration and speed variable, setting an acceleration threshold value When the absolute value of the current acceleration is smaller than the threshold value, the acceleration at the next moment To the maximum acceleration interval, or to the maximum deceleration The interval is randomly selected.

Description

Dangerous driver longitudinal model construction method based on Lewy random characteristics Technical Field The invention belongs to the technical field of intelligent driving tests, relates to a method for constructing a longitudinal model of a dangerous driver, and particularly relates to a method for constructing a longitudinal model of a dangerous driver based on the random Lewy characteristics, which is oriented to intelligent driving assistance ACC and AEB system virtual tests. Background With the gradual upgrade of automatic driving technology, adaptive Cruise Control (ACC) and Automatic Emergency Braking (AEB) are critical longitudinal driving assistance systems, whose reliability and safety verification face a great challenge. The virtual test technology has remarkable advantages in the development and verification of ACC and AEB systems, in particular in the aspects of improving test efficiency, reducing research and development cost, covering edge scenes and the like. At present, dangerous driver model construction is mainly divided into two types, one is to simulate random behaviors of a vehicle based on a random sampling mode so that the dangerous driver model generates more aggressive behavior characteristics, and the other is to construct a dangerous driver behavior decision model based on a series of algorithms, such as a reinforcement learning method, and generate antagonistic interaction behaviors with a tested automatic driving vehicle. In the construction of a following dangerous scene based on natural driving data, a following dangerous scene construction method is disclosed, natural driving data is collected, a following scene is extracted, a Gaussian mixture model is used for fitting front vehicle acceleration distribution, edge acceleration with lower occurrence probability in the natural driving process is sampled, front vehicle behaviors are randomly controlled, and a following dangerous scene test case is generated. The method has the problems that 1, acceleration state transition matrixes under different speeds are required to be counted respectively, then a Gaussian mixture model is used for fitting front vehicle acceleration data under different speeds respectively, the method is complicated and complicated due to excessive fitting times, 2, in order to ensure that front vehicle acceleration data distribution under each speed can be counted, the running data quantity must meet a large scale level, the data acquisition burden is increased, 3, the front vehicle acceleration simulation process determines the acceleration at the next moment according to the fitted distribution, and the theoretical basis of the vehicle running working condition of the relation between the current acceleration and the acceleration at the next moment is lacking. In the test verification of the simulation acceleration capability of a dangerous driver model, a dangerous driver model for testing the acceleration simulation test capability is disclosed, and a dangerous driver behavior decision model generates corresponding control instructions through sensing and analyzing the current vehicle state based on a series of algorithms and parameters. The simulation acceleration capability of the dangerous driver model is verified from two aspects of driving excitation degree of the driver model and active generation capability of the dangerous scene by deploying the model to a joint simulation verification platform. According to the method, simulation duration 2000s is set in verification execution, TTC abnormal threshold is set to be 1s, headway abnormal threshold is set to be 0.6s, occurrence times of dangerous scenes are counted, and the number of times of actively generating dangerous scenes is obviously increased compared with a conventional driver model when the main vehicle, namely, the detected dangerous scenes of the vehicle, namely, abnormal fragments with TTC and headway smaller than set values respectively. The method has the problems that the method is not compared with other dangerous driver models at present, the superiority cannot be embodied, and in addition, the defects of the design of the tested vehicle system cannot be found in time. In the ' unmanned aerial vehicle group resource scheduling method based on improved wolf ' algorithm ', when the global optimal position of the current population is scaled, the method for generating the Laevice random number is utilized, and the Laevice distribution parameter is a default constant value to generate the Laevice distribution random number. The problem with this approach is that when there is a set of actual data with the lyer property, the default constant value is used as the lyer distribution parameter and the generated random number is not one that obeys the desired data characteristics. Specific Lewy distribution parameters are further calculated through a fitting method based on actual driving intensity data, and then random numbers obeyi