NGSIM-based Modeling For Driver’s Lane-change Behavior With Game Theory

Car following and lane-change behavior constitute the driving behavior of a vehicle driver. Because of the high potential causing car accident and severe delay, lane-change behavior has gradually become one of the research hotspot of traffic flow theory.First, symmetric exponential moving average method are used to process the NGSIM trajectory data. By taking the single lane-change behavior at the congestion segment as the research object and dividing it into discretionary lane-change (DLC) and mandatory lane-change (MLC) according to different lane-change intentions, macroscopic characteristics are analyzed statistically. It reveals:1) the number of DLC increases with vehicle flow decreasing; 2) more than half of the DLC is motived by a higher speed; 3) MLC occurs in the middle segment of a collector-distributor lane.Secondly, taking relative velocity as the evaluation index, the data of game process through the lane-change behavior are extracted. Using binary logistic regression model to determine the significant factors of lane-change decision, non-cooperative game and cooperative game models are established separately. The maximum likelihood estimation method is used to calibrate the model parameters. The results indicates that it is reasonable to adopt game theory to explain the mechanism of the decision of a lane-change behavior at congestion segment. Furthermore, it is found that there exists cooperative behavior during the lane-change process.Finally, time and lateral trajectories during the execution phase of lane-change process are further studied. With the time constraints that lane-change stage starting and ending set, the complete single variable behavior parameters are extracted. It is discovered that lane change time obeys logarithmic normal distribution and average time is about 6.8s. Based on the analysis of the significant variables, BP neural network model was built to predict the time of a lane-change process. Moreover, a fifth-order polynomial trajectory fitting model is approached to predict the lateral movements and its goodness of fit is beyond 0.72.