Comparative Study Of Microscopic Traffic Simulation Models Calibration Based On Surrogate Models

The traffic simulation model is an effective tool for traffic management and organization design,which can be used to design and evaluate traffic schemes.Being realistic and effective are preconditions for using simulation models.The efficiency of modeling is affected by the calibration method and the complexity of the model.In order to model efficiently and analyze the traffic condition more effectively,this thesis provides a parameter calibration method and a parameter backwards method based on surrogate models.In addition,two kinds of microscopic traffic simulation software with different degrees of difficulty are calibrated and compared.The main contents and results of the research are as follows:(1)The traffic parameters are obtained from the videos shot by a drone.Based on BP neural network and calibration indexes,the parameter sensitivity analysis of VISSIM and FOSIM was carried out to obtain the key parameters.(2)Constructing Kriging models based on simulation models.The key parameters of VISSIM and FOSIM are calibrated through Kriging models and differential evolution algorithm(DE).The research shows that the results of parameter calibration are effective,indicating the Kriging-DE algorithm is effective and applicable.(3)Comparing DE algorithm and particle swarm optimization(PSO).According to mapping relation provided by Kriging model,a comparative study of DE algorithm and PSO algorithm shows that the calibration speeds of both algorithms are fast,but the accuracy and stability of DE algorithm are better than those of PSO algorithm.Based on the Kriging-DE algorithm,two kinds of simulation software are calibrated and compared.The results show that the calibration effect of VISSIM is better than that of FOSIM in terms of the average travel time,while FOSIM performs better than VISSIM in terms of the frequency distribution of travel time.(4)The calibrated parameters are tested,the traffic condition and the reasons for its changes are analyzed in this study.Based on the Kriging-DE model,parameter combinations with the same simulation target are obtained by fast backward estimation.Also,the simulation test and evaluation are carried out.The results show that: 1)The parameter test results are consistent with the a ctual situation;2)The queue length at the intersection is longer when the desired safe distance between vehicles is increasing and the desired speed of leaving the intersection is decreasing;the travel time will increase as the increase of the desired f ollowing distance and the deceleration,the decrease of the maximum acceleration and the desired speed;3)The relative error between the simulation result and the target value corresponding to the backwards parameter set is less than 5%,suggesting the pr oposed backwards method is effective.In this study,the validity and applicability of the parameter calibration method and the simulation models have been verified.Therefore,the theoretical support can be provided for subsequent related research es.Also,an efficient parameter calibration method and a kind of efficient traffic simulation software are explored for engineering.Further,the reference for traffic management is provided by the conclusions obtained from the analysis of the causes of traffic condition changes.Meanwhile,the parameter backwards estimation method can provide a reverse operation mode for using simulation model to design traffic scheme and give suggestions for better driving behavior.