Research On Traffic Accident Chain Evolution Model And Blocking Strategy Based On Markov Chain Theory

As the core part of Advanced Driver Assistance Systems(ADAS)and an important direction of the development of intelligent vehicle safety technology,automotive collision avoidance system has always been a popular and difficult research question for automobile manufacturers and researchers.At present,researchers at home and abroad have carried out a lot of research work on the theory and technology of collision avoidance intelligent decision-making(including collision risk estimation algorithm and collision avoidance control strategy),but there still remain many shortcomings to be improved and tackled.Taking the automobile collision avoidance intelligent decision-making technology as the research object,the thesis puts forward the new research perspective of “road traffic accident chain”(or Accident-Chain of Road Traffic Incident,A-CRTI).Collision avoidance intelligent decision-making method that could reflect the risk evolution pattern(law)in driving process has been explored from several aspects,including A-CRTI risk state classification method,Markov chain theory based A-CRTI risk state evolution model,and A-CRTI blocking strategy based on the evolution model.The research of the thesis has been funded by the National Natural Science Foundation project “Research on the Dynamic Evolution Law and Blocking Strategy of Road Traffic Accident Chain under Connected Vehicle Environment”.Main contents include:Firstly,according to the evolution characteristics of general road traffic accidents in time dimension,the concept of A-CRTI(i.e.,a series of road traffic events arranged in chronological order and eventually forming accidents or near accidents)is proposed.Based on the A-CRTI concept,a research system of A-CRTI is proposed based on the coupling relationship between traffic events,driving risk status and vehicle motion state using Markov chain theory,which provides a theoretical framework for constructing the longitudinal/lateral A-CRTI evolution model.Taking the deceleration curves of evasive braking in rear-end near crashes as the research object,risk state clustering categories of the evasive braking samples from naturalistic driving data are obtained using deceleration curve spectrum clustering analysis based on dynamic time warping distance and curve changing rate dissimilarity,while fuzzy logic rule extraction is carried out for risk estimation index values at brake initiation under different risk state categories(including collision time TTC,headway time THW,and final distance under emergency braking PICUD,which characterize three kinds of uncertain critical conditions respectively).The fuzzy classification rule of driving risk state based on {TTC,THW,PICUD} is finally obtained.Based on the fuzzy classification rule of driving risk state,a rolling time window-based risk evolution state(Markov state)is proposed for A-CRTI evolution model,and a multi-logistic model-based Markov state transition probability estimation method is proposed considering comprehensive human-vehicle-road-environment influential factors.The longitudinal A-CRTI risk state evolution model is finally constructed based on Markov chain theory.The evolution model is trained and verified by naturalistic driving data,and the results show that the model has good prediction accuracy for each risk state,especially high-risk state(the prediction accuracy rate for high-risk state could reach 90%).Based on the event-driven discreteness and time-driven continuity mixed dynamics presented by A-CRTI,a two-layer discrete-continuous hybrid state model describing the lateral A-CRTI risk state evolution is established based on the continuous observation of motion state of vehicle(such as the speed of the vehicle,the relative distance and relative speed of the main conflicting vehicle,etc.).Vehicle motion sequence data from dangerous lane changing and normal lane changing of typical lateral driving scenario are acquired through driver-in-the-loop simulation platform based on Prescan-Simulink joint simulation,based on which the proposed two-layer lateral A-CRTI evolution model is trained and tested.Results show that the model could accurately distinguish the hazard and normal lane change mode in the horizontal scene,and the average prediction accuracy rate could reach 93.9%.Based on the longitudinal and lateral A-CRTI evolution models,construction method of road traffic accident chain blocking strategy is proposed.Firstly,longitudinal/lateral A-CRTI evolution model libraries are constructed for different longitudinal/lateral scenarios.Then,A-CRTI evolution model matching method is constructed based on SVM classifier and clustering/combination rule method(including remaining/leaving in lane scene classification and specific longitudinal/lateral scenario recognition).Finally,longitudinal/lateral A-CRTI blocking strategies are constructed based on the matched A-CRTI evolutionary model,and evaluation methods of blocking strategies are proposed from three aspects including security,timing and accuracy.Finally,simulation experiments are carried out to verify the proposed A-CRTI blocking strategy,indluding 1)Euro-NCAP test scenario simulation experiment based on Prescan,2)Matlab virtual online simulation experiment based on naturalistic driving data,and 3)typical dangerous lane changing scenario simulation test based on Prescan.Simulation results show that: 1)the longitudinal blocking strategy proposed in this thesis could effectively avoid rear-end accidents when the front vehicle being static,at low speed,and decelerating in Euro-NCAP tests;2)the longitudinal early warning blocking strategy could generate accurate driver warning over one second in advance of the dangerous situation,which could effectively reduce the occurrence of rear-end accidents;3)The early warning timing generated by the lateral warning blocking strategy is 0.78 sec earlier than the NHTSA-recommended warning time at TTC=2.4s,which could effectively reduce the occurrence of lateral lane changing accidents.