Research On Risky Driving Behavior And Vehicle State Prediction Method For Decision Making Of Shared Control Authority

Frequent traffic accidents have proved that driving is a kind of high-risk project,and the risky driving behavior is the main cause of traffic accidents.In recent years,the development of artificial intelligence(AI)technology has greatly promoted the progress of intelligent driving technology.Taking the automatic driving system as the alternative role to assist or replace the human driver is regarded as an effective way to fundamentally solve the safety threat caused by human factors.It is one of the research hotspots in the domain of intelligent vehicle.Facing the shared driving mode in which the driver and the automation system share the control right,when the unnecessary driving task is performed by the driver,which directly or indirectly aggravates the risk of the vehicle.By timely and effectively monitoring the risky status of driver and vehicle,the intelligent vehicle initiates automation in control when the vehicle risk reaches high level,so as to fundamentally curb the potential safety hazards caused by the risky driving behaviors.Based on the in-depth analysis of the research status of related domains,this paper summarizes the problems and deficiencies,determines the main research content and the main research results are as follows:1.In order to solve the problem that the previous research is limited to identify the distracted driving behavior and ignores its impact on driving risk,a time ahead risky driving behavior prediction model based on XGBoost was proposed.Firstly,in order to meet the demands of this research with risk driving behavior as the breakthrough point,an improved NASA-TLX load assessment method suitable for automobile drivers was proposed,which was used to design distracted driving tasks with different types of driving load in the way of precise pressure and enhanced effect.Secondly,by analyzing the abnormal driving state caused by driving tasks and considering its impact on driving performance,the risk level of driving behavior was divided through cross analysis and verification,which provides theoretical support for evaluating the risk level of driving behaviors.Finally,PCA was used to reduce the dimension and integrate the multisource information of vision,physiology and vehicle running state to comprehensively characterize the driving behaviors.After iteratively optimizing the relative optimal time window and other parameters,so as to give full play to the advantages of XGBoost,such as fast operation speed and high prediction accuracy,and a risky driving behavior prediction model based on XGBoost was established.2.In view of the uncertainty caused by some driving risk analysis methods do not consider the differences of drivers' attributes,a vehicle risk driving state prediction method considering drivers' characteristics was proposed to solve the problem of poor generalization ability caused by complex and changeable human factors.Firstly,the structural equation model was used to solve the influencing factors and their weights of braking reaction time(BRT).The parameters of BP neural network were optimized and the performance of the model was improved.By building the prediction method of the BRT,the problem that some human factors are difficult to measure directly was solved and the generalization ability of the model was improved.Secondly,the correlation between lane change duration,self running state,relative running state and driving behavior was analyzed,then the calculation method of lane change duration and longitudinal displacement was proposed,which provides support for the calculation of vehicle trajectory.Finally,the minimum longitudinal safety distance model was used to calculate the minimum longitudinal distance under the premise of no collision between two vehicles,and the response time margin left to the driver of the rear vehicle to avoid collision under the condition of emergency braking of the front vehicle was taken as the driving risk evaluation index.The driving safety in the following scene and merging scene was analyzed,and the driving safety consideration is proposed Vehicle risk driving state prediction method based on driver characteristics.After that,the prediction method of vehicle risk driving state considering driver characteristics was proposed.3.Aiming at the problem of control transition decision and takeover opportunity of the shared driving mode,a decision-making method of control rights based on human vehicle risk status was proposed,which can effectively output reasonable decision results and request automatic takeover in time,which can fundamentally curb the safety hazards caused by risky driving behaviors.Firstly,in order to promote the safety maximization of intelligent vehicles,the risk game model of human and vehicle was established by TOPSIS and complete static game theory.The strategy function of relative utility maximization was proposed and embedded in the reinforcement learning(RL)reward function.Then the RL reward and punishment mechanism guided by the expectation of maximizing vehicle safety was obtained.Secondly,taking advantage of the good performance on solving sequential decision-making problems of RL,a decision-making method of shared driving control based on A2 C algorithm was proposed.The output effect of the decision-making model was optimized by adjusting the weight and reward function of human and vehicle risk decision-making.The effectiveness of the training process and results were verified by using the model performance evaluation indices.Finally,the influence of transition time on vehicle safety was analyzed through simulation test.The decision-making method for shared control authority which could timely and effectively curb the risky driving behaviors and improve the vehicle safety.