Research On Parameters Calibration Of Microscopic Traffic Simulation Model Based On Improved Genetic Algorithm

Traffic simulation has gradually become a mainstream tool for analyzing traffic problems because of its economy,convenience,visibility,and repeatability.The accuracy of the traffic simulation software parameter setting directly affects the quality of the simulation model running results.The operational effectiveness mainly depends on the accuracy of the model,so the parameter calibration of the simulation model is very important.However,most simulation software is developed from abroad.Therefore,it is necessary to recalibrate the model parameters according to the actual local situation,so that make the simulation model closer to the actual situation.This paper studies the parameter calibration of the urban road microscopic traffic simulation model.In the past research,single index or two indicators such as travel time,queue length and speed were selected as evaluation criteria,and various optimization algorithms were used to perform optimal solution search in space,so that the simulation results were closer to the actual observation.However,recent research has shown that according to the traditional calibration method,the calibration result is a combination of countless parameter values,there is a huge solution space,and the optimal solution is the one of the solution spaces,but the traffic characteristic is unique in the actual road.The reason is that different driving behaviors may result in the same travel time or queue length,etc.Only travel time or other index can not map driving behavior.Aiming at this problem,this paper establishes a complete set of traffic running evaluation index system,which integrates the indicators that can reflect the traffic status,so that make the simulation results closer to the real traffic state.At the same time,the comprehensive index of traffic operation status is used as the evaluation index,so that the solution space is greatly reduced,the optimal solution is more reasonable.It is necessary to test the total influence of the changes of multiple parameters on the model operation results.Therefore,the global sensitivity analysis of parameters is based on the comprehensive evaluation index of traffic conditions,thus reducing the number of parameters.In recent years,more and more intelligent algorithms have been used to perform parameter calibration.Among them,genetic algorithm has global optimization ability,computational efficiency and simplicity,and many studies have proved that genetic algorithm can obtain acceptable calibration results or close to global optimal,that genetic algorithm is widely recognized and used.Therefore,the genetic algorithm is used to calibrate the parameters of the micro-simulation model in this paper.The genetic algorithm is easy to fall into the local optimum and the convergence speed is slow.The initial population,selection,crossover and mutation of the algorithm are improved.Improve and strengthen the global search and local search capabilities of the algorithm,thus improving the accuracy and convergence speed of the algorithm.At the same time,the design of the calibration is independent of the characteristics of the traffic model,so it is easy to be used for calibration of other models.The results show that calibration based on the establishment of the traffic operation state evaluation index system is better than based on the single indicator such as travel time,and the goodness of fit test results are significantly improved,that is,the results are more accurate.At the same time,it verifies the importance and necessity of the establishment of the evaluation index system for the parameter calibration work.By improving the genetic algorithm,the convergence time is reduced by 58.8%.Sum up,previous studies have often focused on the selection or hybrid of algorithms,while ignoring the importance of the evaluation indicators for parameter calibration and the improvement of algorithm.Global sensitivity analysis is more comprehensive than local sensitivity analysis,remedying the possibility that local sensitivity analysis may miss or redundant.Improved genetic algorithm has a significant effect on the convergence speed.The establishment of traffic state evaluation indicators,global sensitivity analysis and improved genetic algorithm are designed independently of the traffic model characteristics,so it is easy used for other models calibration.