Research On Decision And Control Algorithm For Vehicle Full-Speed Adaptive Cruise Based On Learning Control

With the continuous progress of urban construction,the originally traffic environment becomes complex and diverse.At the same time,the annual increase of car ownership makes more and more novice drivers plagued with manipulation tasks and decision-making problems.However,although the productization of ACC system on the market can deal with most of driving assistance problems at present,its applicability is limited by the vehicle velocity.Thus,without the consideration of drivers'driving characteristics and vehicles'dynamic characteristics,the ACC products will lack of humanization,which lead to suitability compromised.Therefore,the decision and control algorithm for vehicle full-speed adaptive control based on learning control was researched in this paper.Through the in-depth understanding and analysis of the longitudinal driving rules of real drivers,the humanized decision-making problem of ACC system was converted into the design of learning algorithm based on machine learning theory.In this way,the applicability of the current system will be improved and the function of ACC will achieve to a full-speed ACC with characteristics of drivers'driving behavior.Firstly,when dealing with the problems of drivers'longitudinal driving behavior characteristics,in this paper,several experimental tests were conducted through driving simulator and real vehicle platform for car-following and“stop and go”.In this way,the key parameters were obtained for repersenting the drivers'longitudinal driving behavior.At the same time,based on above key parameters and the combination between safety index and efficiency index,the fuzzy evaluation system for drivers'driving style was established based on Sigmoid function.Moreover,through the comprehensive evaluation,the weighted value R~+and R~-were obtained representing the comprehensive evaluation index,which made more accurate for the classification of drivers'longitudinal driving behavior.Secondly,when dealing with the problems of decision-making and control method for car-following,through the in-depth analysis of drivers'longitudinal behavior,the Markov Decision Process model was established under the mechanism of imitating drivers'car-following behavior,which also revealed the intrinsic relationship between drivers'behavior characteristics and vehicle autonomous decision-making based on the design of reinforcement Q-learning.In this way,the traditional rule-based decision-making method was converted into a learning-based intelligent method.Among them,the state set and the action set were designed based on driving risk principle,the reward functions were designed based on different motion states for vehicle and the value funtion V(s)related to the state was converted into the evaluation function Q(s,a)related to the action through the Bellman Equation.At the same time,the vehicle dynamic characteristics were taken into consideration when establishing vehicle inverse longitudinal dynamics model.Moreover,with the combination between the engine external characteristics and braking system model,the results of decision-making process were converted into detailed control instruction.Then,when dealing with the problems of decision-making and control method for“stop and go”,through the in-depth analysis of drivers'longitudinal behavior for“stop and go”,the method for vehicle automatic driving and braking at low speed was proposed.In the part of driving,the acceleration curve is fitted by multistage fast Fourier transform and the tracking control of expected acceleration was realized based on iterative learning algorithm.In the part of braking,the expected deceleration was calculated by dynamic time to collision under the decision-making mechanism of ideal braking model and the vehicle inverse longitudinal dynamics model and braking system model were combined to realize the tracking control of expected deceleration.At the same time,considering the mutual independence between vehicle driving control and brakingcontrol,this paper designed a mode switching strategy of driving and braking control and optimized it by using the threshold hysteresis value,so as to realize the automatic“stop and go”function for vehicle at low speed.Finally,when dealing with the problems of algorithm verification,a series of simulation tests and real vehicle tests were conducted.Among them,CarSim platform was used to set the simulation environment of vehicle parameters and working conditions and realized the simulation test for car-following and“stop and go”.Moreover,dSPACE MicroAutoBox was used as the external controller of the vehicle,and the designed algorithm was embedded to control the vehicle.The real vehicle tests for car-following and“stop and go”were conducted through the urban expressway condition.Therefore,the full-speed adaptive cruise decision-making and control algorithm was verified by simulation and real vehicle tests.