基于空间域分解的交通网络仿真并行化方法

为了提高交通网络微观仿真的速度,节省仿真运算时间,将并行计算技术应用于微观交通仿真中,建立了交通网络并行仿真模型并在工作站集群上实现.模型中采用了基于空间域分解的并行方法,该方法可以将不同地理位置的交通子网分配到集群中的每台结点机进行仿真运算.通过在5个不同规模路网上的并行仿真实验,对提出的交通并行仿真模型进行了测试.实验结果表明,该并行仿真模型可大大减少交通网络仿真计算时间,与单机仿真相比,并行仿真平均可节省62.9%的运行时间.同时由于随着结点机数量的增加,并行仿真中的通信开销也会相应增加,因此对于特定的交通网络方案及仿真逻辑,存在最优的结点机数量使得仿真运算时间最少.

A Method for Parallelizing Traffic Network Simulation Based on Space Domain Decomposition

In order to speed up traffic network microscopic simulation,the model and implementation of traffic micro-simulation with parallel computing techniques were suggested to reduce the time spent in simulation computation based on a cluster of workstations.The parallelization method is a space domain decomposition,which means that each workstation of the cluster is responsible for a different geographical area of the simulated region.The way how information between the domains is exchanged and how transportation network graph is partitioned were also described.The parallel traffic simulation method was tested on five scenario networks of different size.The computational results indicate that the parallel implementation based on cluster performs better than the sequential computation and the time taken to run simulation is reduced greatly,up to average 62.9% timesaving,though trade-offs between communication(overheads) and computational savings become critical with the number of processors.This suggests that the threshold number of processors for optimal computational performance depends on the specific problem and its solution logic.