SPSA算法在微观交通仿真模型VISSIM参数标定中的应用

Application of Simultaneous Perturbation Stochastic Approximation Algorithm in Parameter Calibration of VISSIM Microscope Simulation Model

微观交通仿真模型在交通系统管理、控制和优化中得到了广泛的应用.然而微观交通仿真模型参数标定是一项复杂且系统的工作,特别是对于较复杂网络,其参数标定耗时长,且不容易找到最优解.本文选取了应用较为广泛的VISSIM仿真模型作为基础平台,针对遗传算法(GA)的不足,建立了基于同步扰动随机逼近(SPSA)算法的微观仿真模型参数标定方法,并实现了程序的自动化标定;最后将该方法应用于北京市快速路仿真模型的驾驶员行为参数标定中,以速度的相对误差平方和作为收敛函数,通过对比GA算法,SPSA算法收敛速度快1.7倍,且在标定后的流量检验中相对误差的平方和小0.16,验证了SPSA算法在VISSIM参数标定上的优越性.

Microscopic traffic simulation models have been widely applied in transportation management, control, and optimization. However, since the calibration of parameters of microscopic traffic simulation models is a complex and sys- tematic process, the time to complete the calibration is usually long and it is difficult to find the optimal solution, espe- cially for the large and complex network. This paper first selects the widely used VISSIM model as the basic platform for the parameter calibration. Then a parameter calibration approach based on simultaneous perturbation stochastic approxi- mation （SPSA） algorithm is proposed and a corresponding automatic calibration procedure is developed. Finally, the proposed approach is applied to the driving behavior parameter calibration of the simulation model for the expressway road network of Beijing, in which the stun of squared relative errors of the speed is used as the objective function. After a comparison of the fitness values of genetic algorithm （GA） and SPSA algorithm, it is shown that the conver- gence of the SPSA algorithm is 1.7 times faster than that of the GA algorithm, and the squared relative errors of traffic volumes using SPSA algorithm are 0.16 smaller than those using GA algorithm, which validates the advan- tage of SPSA algorithm in VISSIM parameter calibration.