CN111582586B - Multi-fleet driving risk prediction system and method for reducing jitter - Google Patents

Multi-fleet driving risk prediction system and method for reducing jitter Download PDF

Info

Publication number
CN111582586B
CN111582586B CN202010394375.6A CN202010394375A CN111582586B CN 111582586 B CN111582586 B CN 111582586B CN 202010394375 A CN202010394375 A CN 202010394375A CN 111582586 B CN111582586 B CN 111582586B
Authority
CN
China
Prior art keywords
vehicle
time
state
driving
driving behavior
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010394375.6A
Other languages
Chinese (zh)
Other versions
CN111582586A (en
Inventor
郝威
刘理
胡林
李焱
龚野
杜荣华
易可夫
王正武
马昌喜
龙科军
吴伟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Changsha University of Science and Technology
Original Assignee
Changsha University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Changsha University of Science and Technology filed Critical Changsha University of Science and Technology
Priority to CN202010394375.6A priority Critical patent/CN111582586B/en
Publication of CN111582586A publication Critical patent/CN111582586A/en
Application granted granted Critical
Publication of CN111582586B publication Critical patent/CN111582586B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0635—Risk analysis of enterprise or organisation activities
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/09—Arrangements for giving variable traffic instructions
    • G08G1/0962—Arrangements for giving variable traffic instructions having an indicator mounted inside the vehicle, e.g. giving voice messages
    • G08G1/0967—Systems involving transmission of highway information, e.g. weather, speed limits
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/16—Anti-collision systems
    • G08G1/166—Anti-collision systems for active traffic, e.g. moving vehicles, pedestrians, bikes
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Abstract

The invention discloses a multi-fleet driving risk prediction system and method for reducing jitter, wherein the prediction method comprises the following steps: acquiring vehicle state characteristic information within a period of time through a data acquisition module; the data analysis module identifies the driving behavior state through a time recursive neural network (LSTM), and determines a final driving behavior state by combining a time lag function; determining a behavior dynamic model of each vehicle according to the driving behavior recognition result, and calculating the actual acceleration of the vehicle through the following dynamic model; and determining the driving risk probability of all relevant vehicles under various driving behavior states based on a Monte Carlo simulation method and an importance sampling strategy. The method combines the time recursive neural network LSTM and the time lag function, can quickly, effectively and stably identify various driving behaviors and driving strategies and evaluate risks, and improves the accuracy of risk evaluation.

Description

Multi-fleet driving risk prediction system and method for reducing jitter
Technical Field
The invention belongs to the technical field of intelligent transportation, and relates to a multi-fleet driving risk prediction system and method for reducing jitter.
Background
In recent years, a multi-vehicle cooperative driving system has received wide attention from most research departments and enterprises as one of main application scenes of a vehicle-road cooperative driving system, and related researches have proved that the cooperative driving system can effectively improve traffic operation efficiency and safety and smoothness of automobile driving.
At present, research and application directions of the fleet cooperative control strategy mainly focus on a simpler fixed single fleet control strategy, and research aims at improving driving safety and stability of a single fleet. In the research of cooperative driving, the optimization of vehicle control stability and driving safety on an automatic control level is mostly emphasized, and the driving behavior modeling and driving state prediction under the condition of mixing various driving behaviors are not researched, so that the risks of the current driving behaviors and strategies are difficult to accurately evaluate. Considering the complexity and variability of the actual driving environment and the respective differences of the objectives of the drivers in the fleet, the conversion of the driving behaviors and the mixing of different driving behaviors can cause the driving state of the vehicle to change greatly (even to be separated from the current fleet), thereby seriously affecting the driving safety of the vehicle and easily causing the confusion of the surrounding driving environment.
The prior art 1 (a method and a system for analyzing driving behaviors of multiple vehicles and early warning danger) discloses that a driving behavior identification model based on a neural network and a variable time window determines the type of the driving behavior, but the time window is a predicted value and is difficult to accurately determine the type of the driving behavior; in the prior art 1, a gaussian model based on an error ratio and a monte carlo method are used for carrying out correlation simulation together, only a submodule introduced by an early warning system is used for theoretical introduction or conception, actual discussion is not carried out, and a correlation calculation model is not refined. Although importance sampling is an important strategy in the monte carlo method to reduce variance, the inventors found in practical studies that two factors, namely operation delay and fluctuation period, are not suitable for representing probability by using a uniformly distributed probability density function. In the prior art 1, a least square method is adopted to carry out parameter estimation on a high-order polynomial model, the identified driving behavior state jitter degree is large, and the prediction precision is reduced to a great extent.
Therefore, under the condition that various driving behaviors are mixed, the multi-fleet driving danger prediction system and method for reducing the jitter, which are provided by the invention, provide a basis for dynamically adjusting the driving strategy of the vehicle and improving the safety and the stability of the driving of the vehicle, and have important significance.
Disclosure of Invention
In order to solve the problems, the invention provides a multi-fleet driving risk prediction method for reducing jitter, which can be used for quickly, effectively and stably identifying various driving behaviors and driving strategies and evaluating risks by combining a time recurrent neural network (LSTM) and a time lag function, so that the risk evaluation accuracy is improved, and the problems in the prior art are solved.
It is another object of the present invention to provide a multi-fleet driving risk prediction system for reducing shudder.
The technical scheme adopted by the invention is that the multi-fleet driving risk prediction method for reducing jitter specifically comprises the following steps of:
s1, acquiring vehicle state characteristic information within a period of time through a data acquisition module, and inputting the information to a data analysis module;
s2, the data analysis module identifies the driving behavior state through a time recursive neural network (LSTM), and determines a final driving behavior state together with a time lag function; determining a behavior dynamic model of each vehicle according to the driving behavior recognition result, and calculating the actual acceleration of the vehicle through the vehicle following dynamic model;
the time-lag function is calculated by equation (1):
Figure BDA0002486879040000021
in the formula, TW dstate Is the duration of the time-lag function,
Figure BDA0002486879040000022
is TW dstate Based on (d), is greater than or equal to>
Figure BDA0002486879040000023
Adjusted value for intensity of tan h function, <' > based on the measured value>
Figure BDA0002486879040000024
Based on a value of the time-lag function, a parameter->
Figure BDA0002486879040000025
And &>
Figure BDA0002486879040000026
The values in each driving behavior state are different, dstate is the driving behavior state determined by the time recursive neural network LSTM by a time lag function method, and the method comprises the following steps: CCF is a cooperative car following state, NCF is a common car following state, ACF is an abnormal car following state, THW is the head time distance between a target car and a previous car, v is the target speed, s is the distance between the target car and the previous car, and safeGap is a user-defined safety distance;
and S3, determining the driving risk probability of all relevant vehicles in various driving behavior states based on the Monte Carlo simulation method and the importance sampling strategy.
Further, in the step S2, the parameters
Figure BDA0002486879040000027
Duration value TW of time-lag function dstate Associated with the headway value THW, defining: />
Figure BDA0002486879040000028
The time-lag function parameter under the abnormal following state is larger than that under other states: />
Figure BDA0002486879040000029
Figure BDA00024868790400000210
Further, in the step S2, time lag function parameters under various driving behavior states are estimated based on a least square method
Figure BDA00024868790400000211
Cost function of least squares, see equation (2):
Figure BDA0002486879040000031
in the formula, dstate i Is a driving behavior state, rstate, determined by a time recurrent neural network LSTM in combination with a time lag function i Is the true state marked as the training data set, diff (dstate) denotes the level of change of the determined driving behavior state, and λ is a weighting factor for comparing the accuracy of the determined driving behavior state and the level of change.
Further, in S3, the driving risk probability P (D) of the relevant vehicle is calculated according to equation (5):
Figure BDA0002486879040000032
wherein B is a driving behavior control pattern, x is a parameter of the driving behavior control pattern of each vehicle in the fleet,
Figure BDA0002486879040000033
is the actual acceleration of the vehicle generated from the driving behavior control pattern of each vehicle in the fleet; />
Figure BDA0002486879040000034
For a time period of state of the associated vehicle, including position, speed and acceleration, o and t represent a vehicle index and a time index, respectively; f (D) represents a dangerous state of the vehicle, and depends on whether the minimum headway of each vehicle and the front vehicle is smaller than a preset headway threshold value or not, and is shown in an equation (4); />
Figure BDA0002486879040000035
Wherein the content of the first and second substances,
Figure BDA0002486879040000036
for a minimum headway in a time zone in relation to a preceding vehicle, in conjunction with a control unit>
Figure BDA0002486879040000037
A predetermined safe headway threshold for the associated vehicle.
Further, the determination of the actual acceleration of the vehicle is specifically as follows: according to a CACC vehicle following dynamic model, a cooperative vehicle following dynamic model with a small fixed headway is adopted, and an OV optimal speed dynamic model in a common vehicle following dynamic model is combined to calculate according to the formula (3):
Figure BDA0002486879040000038
in the formula, acc des (t) is the desired acceleration of the target vehicle at time t, acc act (t) is the actual acceleration of the target vehicle at time t, τ is the vehicle maneuver delay time, Δ t is the simulated time step, β is the time correction factor, acc act (t- Δ t) indicates an actual acceleration of the target vehicle at a certain time point.
Further, the driving risk probability P (D) of the relevant vehicle is calculated based on a Monte Carlo simulation method of an importance sampling strategy, see formula (6);
Figure BDA0002486879040000039
wherein P (X) represents the probability of a dangerous state, X represents a parameter of a driving behavior control pattern of each vehicle in the fleet, P (X) is a random variable probability density function subject to a specific distribution, f (X) is a random sampling result, and f (X) is a probability of a dangerous state i ) Is the random sampling result each time, r is the random simulation times;
further, the driving risk probability P (D) of the associated vehicle is sampled based on the introduced probability density distribution q (x), introducing a sampling weight value w (x) i ) And (3) calculating:
Figure BDA0002486879040000041
where q (x) is an introduction factor with a probability density function, replacing p (x) with q (x) as the sampling function.
Further, the probability density function is based on ascending and descending slope distribution probability density functions, see equations (8) - (9);
Figure BDA0002486879040000042
Figure BDA0002486879040000043
in the formula, a and b are the variation range of a random variable x, and x is an independent variable of q (x) or p (x).
Further, in S1, the vehicle state characteristic information in a period of time includes: the speed difference between the target vehicle and the front vehicle, the acceleration difference between the target vehicle and the front vehicle, the distance between the target vehicle and the center line of the current lane and the head time distance between the target vehicle and the front vehicle within a period of time before the current moment.
A multi-fleet driving risk prediction system for reducing jitter comprises a data acquisition module, a data analysis module, an interaction module and a visualization module;
the data acquisition module is used for acquiring vehicle state characteristic information within a period of time through the vehicle-mounted sensing element and inputting the vehicle state characteristic information to the data analysis module;
the data analysis module comprises a driving behavior state recognition submodule, a following dynamic model submodule and a risk prediction submodule; the state identification submodule is used for identifying the driving behavior state through a time recursive neural network LSTM and determining the final driving behavior state by combining a time lag function;
the following dynamic model submodule is used for determining a behavior dynamic model of each vehicle according to the driving behavior recognition result and calculating the actual acceleration of the vehicle through the following dynamic model;
the risk prediction submodule is used for calculating the driving risk probability of all relevant vehicles under various driving behavior states based on a Monte Carlo simulation method and an importance sampling strategy;
the interaction module is used for selecting a proper prediction result to interact to the visualization module, so that road travel information is provided for a user;
and the visualization module is used for analyzing data and displaying the state evaluation result of the sampling part.
The beneficial effects of the invention are:
(1) The invention provides a multi-fleet driving risk prediction method, which is characterized in that an integrated driving behavior recognition model combining a time recursive neural network (LSTM) and a time-lag function is established under a mixed state of multi-vehicle cooperative driving and various driving behaviors, so that a large amount of unnecessary driving behavior state conversion can be obviously reduced, the recognition result is more stable, the vehicle coverage of risk prediction is favorably reduced, and the calculation efficiency is improved.
(2) The driving risk probability of all relevant vehicles in various driving behavior states is determined based on the Monte Carlo simulation method and the importance sampling strategy, the risk probability can be accurately calculated, the variance of the calculated result of the risk probability is reduced when the risk probability is smaller, the danger and the stability of various driving behaviors of the relevant vehicles can be effectively, accurately and stably comprehensively evaluated, and an important basis is provided for selecting a proper driving strategy.
(3) The invention can also research the driving behavior modeling, identifying and danger predicting method under more complex traffic states (such as the influence of traffic lights, pedestrians and electric bicycles), and lays a certain foundation for the research of improving the stability and safety of vehicle driving in more complex driving states and driving environment changes.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic structural diagram of a prediction system according to an embodiment of the present invention.
FIG. 2 is a driving state recognition and decision diagram of an embodiment of the present invention.
Fig. 3 is a diagram of an experimental scenario in an example of the present invention.
FIG. 4a is a labeled driving behavior category in an example of the invention.
Fig. 4b is the driving behavior recognition result using LSTM prediction.
Fig. 4c is a predicted driving behavior recognition result of an example of the present invention.
Fig. 5a is a result of calculation of the risk probability of the following vehicle when the preceding vehicle is in the abnormal following state in case 1.
Fig. 5b is the calculation result of the risk probability of the following vehicle when the preceding vehicle is in the abnormal following state in case 2.
Fig. 6a is the calculation result of the standard deviation of the risk probability of the following vehicle when the preceding vehicle is in the abnormal following state in case 1.
Fig. 6b is the calculation result of the hazard probability standard deviation of the following vehicle when the preceding vehicle is in the abnormal following state in case 2.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The embodiment of the invention discloses a multi-fleet driving risk prediction method for reducing jitter, which is specifically carried out according to the following steps:
s1, a data acquisition module acquires vehicle state characteristic information within a period of time and inputs the information to a data analysis module; the vehicle state characteristic information includes: the speed difference between a target vehicle and a front vehicle, the acceleration difference between the target vehicle and the front vehicle, the distance between the target vehicle and the center line of the current lane and the head time distance between the target vehicle and the front vehicle within a period of time before the current moment;
s2, the data analysis module identifies the driving behavior state through a time recursive neural network (LSTM), and determines the final driving behavior state by combining a time lag function; determining a behavior dynamic model of each vehicle according to the driving behavior recognition result, and calculating the actual acceleration through the following dynamic model;
the LSTM (Long Short-Term Memory) time recurrent neural network controls the transmission state through the gating state, remembers that unimportant information needs to be memorized for a Long time, can have better performance in a longer sequence and solves the problem of gradient disappearance in the Long sequence training process.
S21, determining the final driving behavior state by combining the time recursive neural network LSTM and the time lag function, where the specific flow is as shown in fig. 2, and FVVD (Front Vehicle behavior Difference) in fig. 2: the speed difference between the target vehicle and the previous vehicle within a period of time (including the current time) before the current time; fVAD (Front Vehicle Accelation Difference): the acceleration difference between the target vehicle and the front vehicle; LP (late Position): the distance between the target vehicle and the current lane center line.
Calculating the time delay function duration by (1) and reducing the vehicle coverage of danger prediction;
Figure BDA0002486879040000061
in the formula, TW dstate Is the duration of the time-lag function,
Figure BDA0002486879040000062
is TW dstate Based on (d), is greater than or equal to>
Figure BDA0002486879040000063
Adjusted value for intensity of tan h function, <' > based on the measured value>
Figure BDA0002486879040000064
Is a reference value of the time lag function value, and a parameter->
Figure BDA0002486879040000065
And &>
Figure BDA0002486879040000066
Values in all driving behavior states are different, dstate is the driving behavior state determined by a time-lag function method through a time recursive neural network LSTM, CCF is a cooperative vehicle following state, NCF is a common vehicle following state, ACF is an abnormal vehicle following state, THW is the head time distance between a target vehicle and a previous vehicle, v is a target vehicle speed, s is the distance between the target vehicle and the previous vehicle, and safeGap is a custom safety distance.
In order to improve the operation speed and avoid falling into a local optimal point, the invention provides some limits on the parameters:
a. the range of the variation of the parameters is,
Figure BDA0002486879040000071
duration value TW of time-lag function dstate Associated with the headway value THW, THW is within 5 seconds in most following states, so that the variation range of corresponding parameters is limited according to the actual following behavior state, and the duration value TW of the time-delay function is limited dstate The basic limitations are as follows:
Figure BDA0002486879040000072
b. the characteristic of behavior state, under unusual car state with, the driving behavior state recognition result can change frequently, for the stability of state recognition result under the unusual car state with, the time lag function parameter under the unusual car state with will be greater than the value under other states:
Figure BDA0002486879040000073
the time-lag function parameters under various driving behavior states are estimated based on the least square method by combining the parameter limits
Figure BDA0002486879040000074
Cost function of least squares, see equation (2):
Figure BDA0002486879040000075
in the formula, dstate i Is a driving behavior state, rstate, determined in combination with a time-recursive neural network LSTM and a time-lag function i Being labeled as the true state of the training data set, diff (dstate) being the difference of the determined states, representing the level of change of the determined driving behavior state, λ being a weighting factor for comparing the accuracy of the determined driving behavior state with the level of change; the formula (2) not only maintains the accuracy of the determined state, but also can obviously reduce a large amount of unnecessary driving behavior state conversion (reduce the jitter in the determined state), so that the identification result is more stable.
S22, according to a CACC vehicle following dynamic model provided by M Segata (Michelle Segata), a cooperative vehicle following dynamic model with a small fixed headway is adopted, and an OV optimal speed dynamic model in a common vehicle following dynamic model is combined to calculate the actual acceleration by the formula (3):
Figure BDA0002486879040000076
in the formula, acc des (t) is the desired acceleration of the target vehicle at time t, acc act (t) is the actual acceleration of the target vehicle at time t, tau is the vehicle control delay time, delta t is the simulation time step length, and delta t is 0.1s in the embodiment of the invention; beta refers to the time correction factor, related to the steering delay time, the sampling interval, acc act (t- Δ t) indicates the actual acceleration of the target vehicle separated from time t by a certain step length;
s3, calculating the driving risk probability of all relevant vehicles in various driving behavior states based on a Monte Carlo simulation method and an importance sampling strategy, and only calculating the risk probability of the vehicle in a certain range behind the abnormal vehicle in the lane where the abnormal vehicle is located for the vehicle in the abnormal following state, so that the calculation efficiency of the risk probability is improved; defining a dangerous state as a state that the distance between a vehicle and a front vehicle is less than a preset headway, setting an expected headway THW to be 1.6 seconds in the embodiment of the invention, and specifically comprising the following steps:
in one simulation, the vehicle danger state f (D) depends on whether the minimum headway of each vehicle O and the front vehicle is smaller than a preset headway threshold value;
Figure BDA0002486879040000081
in the formula (I), the compound is shown in the specification,
Figure BDA0002486879040000082
for a state of the vehicle concerned over a period of time, including position, speed and acceleration>
Figure BDA0002486879040000083
For a minimum headway in a time zone in relation to a preceding vehicle, in conjunction with a control unit>
Figure BDA0002486879040000084
Reservation for associated vehicleThe safe headway threshold, o and t, represents a vehicle index and a time index, respectively.
Driving risk probability P (D) of the relevant vehicle, see equation (5):
Figure BDA0002486879040000085
wherein B is a driving behavior control pattern, x is a parameter of the driving behavior control pattern of each vehicle in the fleet,
Figure BDA0002486879040000086
is a vehicle acceleration generated from a driving behavior control pattern of each vehicle in the fleet; and laying a cushion for calculating the danger probability of multiple vehicles and multiple fleets based on importance sampling and Monte Carlo simulation.
Calculating the driving risk probability P (D) of the relevant vehicle based on a Monte Carlo (Monte Carlo) simulation method of an importance sampling strategy, and obtaining an expression (6);
Figure BDA0002486879040000087
wherein P (X) represents the probability of a dangerous state, X represents a parameter of a driving behavior control pattern of each vehicle in the fleet, P (X) is a random variable probability density function subject to a specific distribution, f (X) is a random sampling result, and f (X) is a probability of a dangerous state i ) Is the random sampling result each time, r is the random simulation times; the Monte Carlo (Monte Carlo) method can solve the problem of determination and then obtain a numerical result through a random sampling algorithm.
The driving risk probability P (D) of the vehicle concerned is sampled on the basis of the introduced probability density distribution q (x), introducing a sampling weight value w (x) i ) Performing calculation, see formula (7), for adjusting and measuring the importance of each sample;
Figure BDA0002486879040000091
wherein q is(x) Is a leading factor with a probability density function (e.g. a uniform distribution function), replaces p (x) with q (x) as a sampling function, samples the weight value w (x) i ) For evaluating the importance of the sampled values using a function q (x) in each sampling simulation; importance sampling changes the sampling distribution of random variables, and allows sampling points to cover points important for calculation results as much as possible within a limited number of sampling times, so that the expectation of the risk probability can be correctly calculated, and the variance of the results is reduced.
Introducing ascending and descending oblique line distribution probability density functions to replace uniformly distributed probability density functions for sampling the importance of related variables;
Figure BDA0002486879040000092
Figure BDA0002486879040000093
in the formula, a and b are x variation ranges of random variables, and x is an independent variable of q (x) or p (x). In the significance sampling simulation, the operation delay is based on a probability density function using a rising diagonal, as shown in equation (8); the period of the fluctuation is based on a probability density function using the falling diagonal, as shown in equation (9).
And simulating the running states of the relevant vehicles within a period of time for many times according to the provided driving behavior dynamic model and the importance sampling strategy, calculating the minimum headway of each vehicle, judging whether the corresponding vehicle and the motorcade are dangerous or not, and finally calculating the weight value of each simulation and the dangerous probability of the relevant vehicle and the motorcade according to the formulas (4) and (7).
The embodiment of the invention discloses a multi-fleet driving risk prediction system for reducing jitter, which comprises a data acquisition module, a data analysis module, an interaction module and a visualization module;
the data acquisition module is used for acquiring vehicle state characteristic information within a period of time through the vehicle-mounted sensing element, acquiring traffic parameters at key positions and inputting the traffic parameters to the data analysis module; the data acquisition module comprises a vehicle-mounted sensing element for acquiring the speed, the acceleration and the position of the vehicle;
the data analysis module is used for processing and analyzing the acquired data and processing the data by adopting a weighted average algorithm; the data analysis module comprises a driving behavior state identification submodule, a following dynamic model submodule and a risk prediction submodule; the state identification submodule is used for identifying the driving behavior state through a time recursive neural network (LSTM) and determining the final driving behavior state by combining a time lag function;
the vehicle following dynamic model submodule is used for determining a behavior dynamic model of each vehicle according to the driving behavior recognition result and calculating the actual acceleration of the vehicle through the vehicle following dynamic model; specifically, according to a CACC vehicle following dynamic model, a cooperative vehicle following dynamic model with a small fixed headway is adopted, and an OV optimal speed dynamic model in a common vehicle following dynamic model is combined to calculate the actual acceleration;
the risk prediction submodule is used for calculating the driving risk probability of all related vehicles under various driving behavior states based on a Monte Carlo simulation method and an importance sampling strategy;
the interaction module is used for selecting a proper prediction result and interacting the prediction result to the visualization module so as to provide road travel information for the user;
and the visualization module is used for analyzing the data and displaying the state evaluation result of the sampling part.
The embodiment of the invention effectively collects the related traffic parameters and lays a foundation for the data analysis module; the NGSIM database and SUMO (urban traffic simulation software) are adopted as a driving behavior data source and a simulation experiment platform, in a simulation experiment, all vehicles in a motorcade can acquire related information of front vehicles and head vehicles of the motorcade in the motorcade, wherein the related information comprises positions, speeds, accelerations and the like, and basic parameters of the vehicles in the experiment are as shown in a table 1:
TABLE 1 vehicle basic parameters
Range of variation of acceleration -7m/s 2 ～4m/s 2
Vehicle handling delay 0 to 1s (even distribution or ascending oblique line distribution)
Vehicle state transmission delay 0 to 1s (even distribution or ascending oblique line distribution)
Maximum speed 40km/h
Random fluctuation coefficient of acceleration 0.05
Random error coefficient for vehicle state observation 0.05
Length of vehicle 6m
Minimum pitch 1m
Maximum fleet length 8 vehicles
The random error coefficient of the vehicle state observation refers to uniformly distributed random errors of which the maximum amplitude of the speed, the acceleration and the distance between the target vehicle and the front vehicle is plus or minus 5% of the current observation value, the random fluctuation coefficient of the acceleration refers to uniformly distributed random errors of which the actual execution acceleration has plus or minus 5%, and the parameters of the driving behavior dynamic model adopt default values.
According to the method, the target headway distance to a front vehicle is set to be 1s under the cooperative vehicle following state and the common vehicle following state, all vehicles are in the cooperative vehicle following or common vehicle following state, and the vehicles in a fleet are in the cooperative driving behavior state. In fig. 3, when the vehicle B is in an abnormal driving behavior state, in case 1, the vehicle 3 behind the vehicle B still maintains a cooperative driving behavior state, and a cooperative following behavior dynamic model is adopted; under the condition 2, the rear 3 vehicles of the vehicle B form a new vehicle fleet, the vehicle C is a head vehicle, a common following behavior dynamic model is adopted, the rest 2 vehicles still keep a cooperative driving behavior state, and a cooperative following behavior dynamic model is adopted. In a simulation experiment, when the vehicle B is in abnormal driving behavior, cosine waves with the period of 4-10s, the average speed of 25km/h and the amplitude of 5-15km/h are randomly changed.
Experimental and results analysis, fig. 4a to 4c show a piece of data for identifying a vehicle driving behavior state identification model, where the ordinate of the graph is the cooperative following state CCF =1, the normal following state NCF =2, and the abnormal following state ACF =3; as shown in fig. 4a, the cooperative following state of the vehicle B is switched from 40-60s to the normal following state, and the cooperative following state of the vehicle B is switched from 80-100s to the abnormal following state. In the invention, the number of LSTM hidden layers is 100, and the data time length of the behavior characteristics is 2s.
As can be seen from fig. 4b, the LSTM model can effectively detect various driving behavior states, but obviously a large number of state transitions with short duration occur, and a considerable portion of the state transitions are frequent, obviously due to the short time and jerky operation of the vehicle in the current 3; as can be seen from fig. 4c, the LSTM combined with the time lag function method can significantly reduce the frequent switching of the LSTM recognition result.
Table 2 shows a comparison of the driving behavior state recognition accuracy and the state change rate of the integrated NGSIM database and the simulation experimental data (obtaining the cooperative vehicle-following behavior data);
TABLE 2 recognition results of driving behavior
Figure BDA0002486879040000111
Compared with the LSTM model, the driving behavior state identification model combining the LSTM and the time-lag function in the embodiment can reduce the times of identifying unnecessary driving behavior state changes under the condition of ensuring the identification accuracy, thereby improving the accuracy and the stability of the dangerous probability calculation of the corresponding vehicle.
In the importance sampling simulation process, vehicle control delay and vehicle state transmission delay time are subjected to importance sampling simulation based on a rising oblique line probability density distribution function, the cycle of a vehicle B in an abnormal vehicle following state is simulated based on a falling oblique line probability density distribution function, and in the uniform distribution sampling simulation process, all variables are subjected to multiple times of simulation based on uniform distribution sampling.
Fig. 5a to 5B and fig. 6a to 6B show the calculation results of the risk probability of the following vehicle using different driving strategies according to the recognition result when the preceding vehicle is in the abnormal following state, and fig. 5a (corresponding to case 1 in fig. 3), fig. 5B (corresponding to case 2 in fig. 3), fig. 6a (corresponding to case 1 in fig. 3), and fig. 6B (corresponding to case 2 in fig. 3) respectively calculate the risk probability of the following 3 vehicles of the vehicle B. In fig. 5a-5b, the importance sample and the uniform distribution sample converge to very close hazard probabilities as the number of simulations increases, whether the C vehicle is in a cooperative driving behavior state or a normal following behavior state, but fig. 6a-6b show that the variance of the importance sample is significantly lower than the method of the uniform distribution sample, indicating that the more stable the hazard probability estimate of the sample is and the less the fleet fluctuates as the value of the variance is smaller. In addition, when a new vehicle fleet is formed behind an abnormal vehicle, the vehicle fleet is safer than the vehicle fleet left behind.
The risk probability estimation of importance sampling adopted in the embodiment is superior to that under uniform distribution sampling, and is reflected in good stability, high convergence speed and low risk probability; the experimental result shows that when the vehicle is in an abnormal following state, the driving behavior identification and risk assessment system provided by the invention can efficiently, accurately and stably identify the vehicle state and assess the risk and the stability of various driving behaviors of the relevant vehicle, and provides an important basis for selecting a suitable driving behavior and strategy.
In the invention, the complicated variability of the actual driving environment and the different purposes of the drivers in the motorcade are considered, the conversion of the driving behaviors and the mixing of different driving behaviors can cause the great change of the driving state of the vehicle, even the vehicle needs to be separated from the current motorcade, thereby seriously affecting the driving safety of the vehicle and easily causing the confusion of the surrounding driving environment. The integrated driving behavior recognition model combining the LSTM and the time-lag function is provided, the driving risk probability of all relevant vehicles in various driving behavior states is calculated based on the Monte Carlo simulation method and the importance sampling strategy, a large amount of unnecessary driving behavior state conversion can be obviously reduced, the recognition result is more stable, and therefore the vehicle coverage range of risk prediction is favorably reduced, and the calculation efficiency is improved.
The above description is only for the preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention shall fall within the protection scope of the present invention.

Claims (8)

1. A multi-fleet driving risk prediction method for reducing jitter is characterized by specifically comprising the following steps:
s1, acquiring vehicle state characteristic information within a period of time through a data acquisition module, and inputting the information to a data analysis module;
s2, the data analysis module identifies the driving behavior state through a time recursive neural network (LSTM), and determines a final driving behavior state together with a time lag function; determining a behavior dynamic model of each vehicle according to the driving behavior recognition result, and calculating the actual acceleration of the vehicle through the vehicle following dynamic model;
the time-lag function is calculated by equation (1):
Figure FDA0004117601160000011
in the formula, TW dstate Is the duration of the time-lag function,
Figure FDA0004117601160000012
is TW dstate Base value of (a), based on>
Figure FDA0004117601160000013
Is an intensity adjustment value of the tanh function>
Figure FDA0004117601160000014
Based on a value of the time-lag function, a parameter->
Figure FDA0004117601160000015
And &>
Figure FDA0004117601160000016
The values in each driving behavior state are different, dstate is the driving behavior state determined by the time recursive neural network LSTM by a time lag function method, and the method comprises the following steps: CCF is a cooperative car following state, NCF is a common car following state, ACF is an abnormal car following state, THW is the head time distance between a target car and a previous car, v is the target speed, s is the distance between the target car and the previous car, and safeGap is a user-defined safety distance;
s3, determining the driving risk probability of all relevant vehicles in various driving behavior states based on a Monte Carlo simulation method and an importance sampling strategy;
in S3, the driving risk probability P (D) of the relevant vehicle is calculated according to equation (5):
Figure FDA0004117601160000017
wherein B is a driving behavior control pattern, x is a parameter of the driving behavior control pattern of each vehicle in the fleet,
Figure FDA0004117601160000018
is the actual acceleration of the vehicle generated from the driving behavior control pattern of each vehicle in the fleet; />
Figure FDA0004117601160000019
For a time period of state of the associated vehicle, including position, speed and acceleration, o and t represent a vehicle index and a time index, respectively; f (D) represents a dangerous state of the vehicle, and depends on whether the minimum headway of each vehicle and the front vehicle is smaller than a preset headway threshold value or not, and is shown in an equation (4);
Figure FDA00041176011600000110
wherein the content of the first and second substances,
Figure FDA00041176011600000111
for a minimum headway of the vehicle concerned from the preceding vehicle over a period of time, is/are>
Figure FDA00041176011600000112
A predetermined safe headway threshold for the associated vehicle;
calculating the hazard probability of multiple vehicles and multiple teams by a Monte Carlo simulation method based on an importance sampling strategy, see formula (6);
Figure FDA0004117601160000021
/>
wherein P (X) represents the probability of danger of multiple vehicles and multiple fleets, X represents the parameter of the driving behavior control mode of each vehicle in the fleets, and P (X) is a random variation subject to a specific distributionThe function of the probability density of the quantity, f (x) is the result of random sampling, f (x) i ) Is the result of each random sampling, and r is the random analog number.
2. The method of claim 1, wherein in step S2, the parameters are used to predict driving risk of multiple fleets for reducing jitter
Figure FDA0004117601160000022
Duration value TW of time-lag function dstate Associated with the headway value THW, defining: />
Figure FDA0004117601160000023
The time-lag function parameter under the abnormal following state is larger than that under other states: />
Figure FDA0004117601160000024
3. The method for predicting driving risk of multi-fleet vehicles for reducing jitter according to claim 1 or 2, wherein in the step S2, the time-lag function parameters under various driving behavior states are estimated based on the least square method
Figure FDA0004117601160000025
Figure FDA0004117601160000026
Cost function of least squares, see equation (2):
Figure FDA0004117601160000027
in the formula, dstate i Is a driving behavior state, rstate, determined by a time-recursive neural network LSTM in combination with a time-lag function i Is the true state marked as the training data set, diff (dstate) denotes the determinationIs a weighting factor for comparing the determined driving behaviour state with the accuracy of the level of change.
4. A multi-fleet driving risk prediction method for reducing shudder as set forth in claim 1, wherein the determination of the actual acceleration of the vehicle is specifically: according to a CACC vehicle following dynamic model, a cooperative vehicle following dynamic model with a small fixed headway is adopted, and an OV optimal speed dynamic model in a common vehicle following dynamic model is combined, and the calculation is carried out by the formula (3):
Figure FDA0004117601160000028
in the formula, acc des (t) is the desired acceleration of the target vehicle at time t, acc act (t) is the actual acceleration of the target vehicle at time t, τ is the vehicle maneuver delay time, Δ t is the simulated time step, β is the time correction factor, acc act (t- Δ t) indicates an actual acceleration of the target vehicle at a certain time point.
5. A multi-fleet driving risk prediction method for reducing jitter according to claim 1, wherein the sampling is based on an introduced probability density distribution q (x), introducing a sampling weight value w (x) i ) And (3) calculating:
Figure FDA0004117601160000029
where q (x) is an introduction factor with a probability density function, replacing p (x) with q (x) as the sampling function.
6. The method of claim 1 or 5, wherein the probability density function is based on ascending and descending diagonal distribution probability density functions, see equations (8) - (9);
Figure FDA0004117601160000031
Figure FDA0004117601160000032
/>
in the formula, a and b are the variation range of a random variable x, and x is an independent variable of q (x) or p (x).
7. The method of predicting driving risk of multi-fleet vehicles for reducing shudder as set forth in claim 1, wherein S1, the vehicle state characteristic information over a period of time comprises: the speed difference between the target vehicle and the front vehicle, the acceleration difference between the target vehicle and the front vehicle, the distance between the target vehicle and the center line of the current lane and the head time distance between the target vehicle and the front vehicle within a period of time before the current moment.
8. The system of claim 1, comprising a data collection module, a data analysis module, an interaction module, a visualization module;
the data acquisition module is used for acquiring vehicle state characteristic information within a period of time through the vehicle-mounted sensing element and inputting the information to the data analysis module;
the data analysis module comprises a driving behavior state recognition submodule, a following dynamic model submodule and a risk prediction submodule; the state identification submodule is used for identifying the driving behavior state through a time recursive neural network (LSTM) and determining the final driving behavior state by combining a time lag function;
the following dynamic model submodule is used for determining a behavior dynamic model of each vehicle according to the driving behavior recognition result and calculating the actual acceleration of the vehicle through the following dynamic model;
the risk prediction submodule is used for calculating the driving risk probability of all relevant vehicles under various driving behavior states based on a Monte Carlo simulation method and an importance sampling strategy;
the interaction module is used for selecting a proper prediction result to interact to the visualization module, so that road travel information is provided for a user;
and the visualization module is used for analyzing data and displaying the state evaluation result of the sampling part.
CN202010394375.6A 2020-05-11 2020-05-11 Multi-fleet driving risk prediction system and method for reducing jitter Active CN111582586B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010394375.6A CN111582586B (en) 2020-05-11 2020-05-11 Multi-fleet driving risk prediction system and method for reducing jitter

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010394375.6A CN111582586B (en) 2020-05-11 2020-05-11 Multi-fleet driving risk prediction system and method for reducing jitter

Publications (2)

Publication Number Publication Date
CN111582586A CN111582586A (en) 2020-08-25
CN111582586B true CN111582586B (en) 2023-04-18

Family

ID=72110761

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010394375.6A Active CN111582586B (en) 2020-05-11 2020-05-11 Multi-fleet driving risk prediction system and method for reducing jitter

Country Status (1)

Country Link
CN (1) CN111582586B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113034911B (en) * 2020-12-14 2022-06-14 湖南大学 Vehicle queue control method and system with parameter and structure heterogeneity
CN113485125B (en) * 2021-08-11 2024-01-30 长安大学 Time-lapse-containing vehicle queue stability control method and system suitable for arbitrary communication topology
CN114118633B (en) * 2022-01-26 2022-06-10 之江实验室 Index self-optimization prediction method and device based on precedence relationship
CN116434603A (en) * 2022-12-05 2023-07-14 东南大学 Automatic driving fleet horizontal and vertical synchronous safety control method based on SSM
CN116386387B (en) * 2023-04-19 2024-03-08 长安大学 Method and device for predicting following behavior of driving vehicle of hybrid queue person

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105448080A (en) * 2015-11-16 2016-03-30 北京理工大学 Modeling method considering influence of sub-adjacent vehicles to traffic flow time lag car-following model stability
CN105612091A (en) * 2013-10-10 2016-05-25 日立汽车系统株式会社 Vehicle motion control device
CN108428343A (en) * 2018-05-17 2018-08-21 长沙理工大学 A kind of more vehicle driving behavior analysis and danger early warning method and system
CN110675656A (en) * 2019-09-24 2020-01-10 华南理工大学 Intelligent vehicle lane change early warning method based on instantaneous risk identification

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2008068832A1 (en) * 2006-12-04 2008-06-12 Fujitsu Limited Driving simulation evaluating method, driving simulation evaluating device and computer program
EP3239686A1 (en) * 2016-04-26 2017-11-01 Walter Steven Rosenbaum Method for determining driving characteristics of a vehicle

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105612091A (en) * 2013-10-10 2016-05-25 日立汽车系统株式会社 Vehicle motion control device
CN105448080A (en) * 2015-11-16 2016-03-30 北京理工大学 Modeling method considering influence of sub-adjacent vehicles to traffic flow time lag car-following model stability
CN108428343A (en) * 2018-05-17 2018-08-21 长沙理工大学 A kind of more vehicle driving behavior analysis and danger early warning method and system
CN110675656A (en) * 2019-09-24 2020-01-10 华南理工大学 Intelligent vehicle lane change early warning method based on instantaneous risk identification

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
惠飞 ; 彭娜 ; 景首才 ; 周琪 ; 贾硕 ; .基于凝聚层次的驾驶行为聚类与异常检测方法.计算机工程.2018,(第12期),全文. *

Also Published As

Publication number Publication date
CN111582586A (en) 2020-08-25

Similar Documents

Publication Publication Date Title
CN111582586B (en) Multi-fleet driving risk prediction system and method for reducing jitter
CN107742193B (en) A kind of driving Risk Forecast Method based on time-varying state transition probability Markov chain
CN107346460B (en) Following operating mode Forecasting Methodology based on the lower front truck operation information of intelligent network contact system
CN109871599B (en) Unmanned vehicle operation safety index evaluation method
da Rocha et al. Does traffic-related calibration of car-following models provide accurate estimations of vehicle emissions?
CN109002622B (en) Method for estimating overall load response of large-span bridge under random traffic flow effect
CN110796859A (en) Real-time traffic state identification and accident risk early warning method based on traffic flow
CN114783183A (en) Monitoring method and system based on traffic situation algorithm
Song et al. Comparative analysis of car-following models for emissions estimation
Bareket et al. Methodology for assessing adaptive cruise control behavior
CN111688713B (en) Driving behavior analysis method and device
CN112466119A (en) Method and system for predicting vehicle following speed of vehicle by using vehicle-road cooperative data
CN111951430B (en) Vehicle drivability evaluation method and system
Gupta et al. Estimation of fuel economy on real-world routes for next-generation connected and automated hybrid powertrains
Astarita et al. Calibration of a new microsimulation package for the evaluation of traffic safety performances
Asamer et al. Calibrating VISSIM to adverse weather conditions
CN113436432A (en) Method for predicting short-term traffic risk of road section by using road side observation data
WO2024021871A1 (en) Vehicle-road collaboration-based sensed roadside test data quality evaluating method
CN115422747A (en) Method and device for calculating discharge amount of pollutants in tail gas of motor vehicle
CN113535816A (en) Driving performance evaluation method and system for intelligent network cloud control vehicle
Archer Developing the potential of micro-simulation modelling for traffic safety assessment
CN109782751A (en) Method and apparatus for autonomous system performance and benchmark test
CN113570852B (en) High-risk road section risk prevention and control means evaluation method based on safety substitution evaluation index
CN113592221A (en) Road section risk dynamic evaluation method based on safety substitution evaluation index
CN114155742B (en) Method for evaluating and early warning longitudinal driving risk of internet vehicle

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant