Research On Lane-changing Prediction And Control Based On Natural Driving Data

Lane change behavior is one of the most basic driving operations,and has received much attention in the automotive and transportation fields.The purpose of changing lanes is to improve or switch driving conditions,but it also poses a huge hidden danger to driving safety.Therefore,in the field of advanced assisted driving and unmanned driving,lane change behavior is regarded as one of the key research objects.In this study,the lane change behavior is taken as the object,and the lane-changing trajectory prediction and trajectory control are studied from the perspective of human-vehicle interaction and road-vehicle interaction respectively,which provides a basis for the establishment of intelligent vehicle driving model.The main research contents are as follows:Firstly,in order to study the lane change behavior characteristics,a natural driving data acquisition experiment was designed.In the experiment,the natural driving data on the designated elevated section are obtained by image measurement Based on the experimental data,the difference between left and right lane change,the relationship between lane change speed and vehicle spacing,and the collision time are analyzed and studied.Secondly,the deep neural network is applied to predict the lane change trajectory.In this dissertation,a trajectory prediction model is designed,which combines long Short-Term Memory(LSTM)neural network with binary Gauss Mixture Model(GMM).In the model,LSTM network is used to perceive and analyze the driving status of object vehicle and surrounding vehicles,and the results are transmitted to the Gauss mixture model to predict the probability distribution of lane change trajectory.Compared with the traditional trajectory prediction model,besides the object vehicle state,the surrounding traffic factors are taken into consideration.Moreover,the output of the model is a continuous probability density distribution,which provides more information.In this dissertation,the model is trained and validated by using the lane change samples in the natural driving data acquisition experiment.The results show that the model has good prediction accuracy.Thirdly,the trajectory plan and control based on data-driven method are designed.This method takes "active" and "passive" lane change as objects,and designs a hierarchical algorithm with natural driving data.The upper-level algorithm combines machine learning method,using Adaboost classifier to make decision on the type of lane change,and transmits the decision results to the middle-level.According to the lane change driving habit,the middle-level algorithm uses polynomial curve method to plan the lane change trajectory in Frenet coordinate system.Lastly,the lower-level algorithm uses the non-linear model predictive control method to make the vehicle travel according to the optimal planning trajectory.Finally,in order to verify the designed trajectory planning and control model,simulation experiments and real vehicle comparison experiments are designed in this dissertation.In the simulation experiment,the trajectory planning and control model is established by MATLAB and Carsim software platform,and two scenarios are set up to verify the model.Then,in order to further verify the trajectory planning effect,a real vehicle comparison experiment is designed.On the premise of the same lane-changing environment,the lane-changing model is compared with the driver by comparing the lane-changing duration and the cumulative longitudinal and lateral vibration.The experimental results show that the model has good trajectory planning effect and tracking control accuracy.The research results in this dissertation provide new methods and ideas for the study of lane change behavior and play an important role in improving both the advanced driver assistance system and unmanned driving system.