Research On Vehicle Trajectory And The Characteristics Of Lane-changing Behavior

The models of car-following and lane-changing are integral part of microscopic traffic flow theory and focus of AI technology in the futher.Although the lane-changing model is relative immature compared with the car-following model,it is a hot research area for thousands of scholars.One of the most difficult part in the research microscopic traffic model is data collection.In the most of previous research,GPS data,video-based data,and simulation data were adapted.However,these data,to some extent,have weakness in the aspect of quantity and quality.NGSIM vehicle trajectory data has brought tremendous break-through as a precise and comprehensive dataset for microscopic traffic model exploration.While,it cannot be used for calibration and validation directly,because many outliers and measurement errors exist in raw NGSIM data.Based on NGSIM vehicle trajectory data,this thesis aims to explore the algorithm for trajectory reconstruction and its application in lane-changing behavior research:Firstly,a two-step algorithm for vehicle trajectory reconstruction was developed:?The two patterns of outliers were identified by wavelet analysis and physical restricts respectively,and re-estimated by Lagrange 5^th-degree and 3th-degree polynomial interpolation;?The Kalman filter,considering signal energy,was conducted to filter the noise in NGSIM data.To demonstrate the feasibility and display the performance of this algorithm,this thesis took“Vehicle ID”1882 as an example,and the result indicated that the algorithm performs well.Then this algorithm was conducted to reconstruct the longitudinal and lateral vehicle trajectories of all the 1942 private cars in NGSIM sub-dataset.Secondly,extracting lane-changing trajectory.This thesis extracted the discretionary and mandatory lane-changing trajectory?DLC&MLC?from the reconstructed NGSIM vehicle trajectory,and using K-means cluster identified four kinds of failure lane-changing trajectory:turn to the current lane from the target lane,ride on the lane line for a long time,change lane frequently in short time,record error in“Lane ID”.Finally,this thesis collected 119 and 45successful DLC and MLC trajectories respectively.Finally,research on the lane-changing behavior.This thesis mainly studied the duration of lane-changing and fitting of the lateral vehicle trajectory of lane-changing,according to lane-changing styles and directions.On the part of duration of lane-changing,this thesis established the rule-based method and set constraint for the two special situation to extract and analysis the duration of DLC and MLC,and left-toward lane-changing and right-toward lane changing systematically.On the aspect of fitting of lateral trajectory of lane-changing,this thesis,regarding the mean absolute error?MAE?,the average root mean square error?RMSE?,and the average root mean square percentage error?RMSRE?as indices,explored the polynomial fitting for lateral trajectory of lane-changing.The results indicated that 5^th-degree polynomial fitting is advisable for DLC and right-toward MLC fitting,and 4^th-degree polynomial fitting is advisable for left-toward MLC fitting.