The Modelling And Simulation Of Traffic Flow At Non-signalized Intersections Based On Game Learning

Non-signalized intersections are places where accident occurs frequently and are important reasons of traffic jam.At non-signalized intersections,vehicles moving at different directions have conflicts about the right to go through,which directly influences traffic safety and traffic system efficiency.To address this problem,many researchers view this phenomenon as a game theory problem,and study traffic flow by modelling drivers' behaviour.However,there isn't much work which considers learning into this background.This paper will use modelling and simulation to study the effect of the learning mechanism on traffic flow on non-signalized intersection system.This thesis will adopt belief-based learning and logistic regression to model game behaviour at non-signalized intersections.In the belief-based model,cooperation rate is derived mathematically and is a function of cost-to-benefit ratio r and memory capacity m,the cooperation rate shows staircase like changes due the change of r and the number of staircases are related to m.In the logistic-regression model,independent variables are strategy chosen by the player and its opponent in the foregoing two round.We use data from experiment to calibrate the parameters.Using cellular automaton traffic flow model,the two behaviour models are simulated in a road network composed of non-signalized intersections.This thesis finds that in the belief-based learning model,memory capacity is useful for drivers to understand the system cooperation rate,thus decreases the possibility of plunge and increases the efficiency of traffic system.This thesis also finds that cooperation is helpful in increasing traffic flow and average speed.Furthermore,depending the total number of vehicles,there are three phase traffic flow phase that have different character.In the logistic regression part,because the parameters in the model has been calibrated using experiment data,the simulation result shows high consistency with experiment data,and this implies that the fitted model has high validity.Besides,the research finds that based on this regression model,on average,the defector always gets less than cooperatorFew past studies on on-signalized intersection consider learning mechanism.This thesis introduces belief-based learning and logistic regression into this scene and studies the effect of drivers past experience on traffic system,which has some theoretical and practical value.