Research On Driving Style Of Car-following And Human-centered Adaptive Cruise Control

In recent years,intelligent driving technology has boomed.According to the intelligence level of the vehicle,the intelligent driving technology can be divided into L1-L5,and the L2 and L3 levels are also called Advanced Driver-assistance System(ADAS).In development of ADAS,there will exist vehicles of different intelligence levels running on the road.Therefore,ADAS should understand human driving behavior,and the driving behavior of the intelligent vehicle equipped with ADAS should be similar to that of human,so that it can be acceptable to human and human-centered control is realized.Adaptive Cruise Control(ACC)is a key technology among ADAS.In order to achieve human-centered control,the demands of drivers with different driving styles should be taken into consideration.Human-centered ACC needs to recognize drivers' car-following style with accuracy,imitate the car-following style with safety and correct improper carfollowing behavior simultaneously.Aiming at these demands,a human-centered adaptive cruise control strategy is proposed in this paper.The strategy can recognize driving style according to car-following behavior,adjust ACC parameters to adapt to driving style and realize human-centered control.Besides,it can optimize the economy in the car-following process properly.The problem about extracting car-following behavior,and driving style classification,car-following speed control and human-centered adaptive cruise control strategy and simulation verification is mainly studied in this dissertation.The highlights and conclusions are as follows:1.Car-following parameters calibrate and driver style classify.Firstly,the driving data in I-80 highway from 4:00pm-4:15pm of NGSIM dataset is chosen and car-following data is extracted from it with programming method.A Symmetric Exponential Moving Average method is used to denoise and filter the car-following data and A set of 1152 drivers' data is obtained.Then,Gipps model is selected to identify car-following behavior from the timeseries data and genetic algorithm is used to calibrate Gipps model's parameters of every driver.Finally,these parameters are clustered by DBSCAN clustering.Depending on Silhouette coefficient,drivers are clustered into three types including risk,neutral and conservative type and a driving style classifying model is established.2.Speed prediction model considering driving styles.Firstly,a four-layer BP neural network is built and optimal topology of 5-5-5-1 and hyperparameters of the network is chosen by using TensorFlow.Then,in order to eliminate the overfitting output speed,a Gipps model speed is introduced and a combined speed control model is built,which modifies the overfitting speed and promotes adaptability to more scenarios.Finally,on the premise of car-following safety,car-following parameters of drivers with different styles are taken into consideration in the speed prediction model,and the optimal weighting coefficients of each style in the speed model are identified,and a speed control model considering driving styles is established.3.Human-centered adaptive cruise control strategy and simulation verification.Firstly,an ACC structure is studied and decision-making and control layer are designed.Secondly,according to the distribution of the drivers' time headway,the boundary speed of drivers with different styles is derived based on the conventional ACC strategy and the economy speed and the safety speed are calculated.The weight coefficients of different speeds are selected to make the output speed of the decision-making layer similar to the driver's style.Then,depending on reverse vehicle longitudinal dynamics model,the control layer is built and follows the reference speed accurately and a simulation platform is built using Carsim and Matlab/Simulink.Finally,four typical car-following conditions and WLTC condition are selected as simulation scenarios and simulation tests are carried out.The results indicate that the proposed human-centered control strategy can achieve speed follow and recognize driving styles accurately,and the control is similar to driver's style.Moreover,according to the reference speed,the economy speed is obtained and the car-following speed can be optimized to improve following economy within driver's acceptance range.