一种基于多Agent的人群仿真多机并行平台研究

Research on Parallel Execution Platform for Multi-Agent-Based Crowd Simulation

本文提出了一种针对多Agent人群仿真模型的多机并行平台,该平台利用MPI作为节点间的通讯软件,采用主从模式设计多机并行执行环境。在主控节点中设置结果回收模块与监控模块,通过K-means算法保证执行节点之间的负载均衡,同时以此作为任务分发与调度的依据;在执行节点(从节点)中,则根据主控节点的任务调度执行相关Agent模型,并执行仿真同步与状态信息汇报等任务。实验结果表明,该多机并行平台能够执行基于多Agent的人群仿真模型,能有效地提高仿真的性能,同时具有良好的可扩展性。

This paper proposes a multi-agent based parallel platform for simulating virtual crowds. It applies MPI as the communication software amongst nodes in master-slave mode. Using the k-means al- gorithm, the master node ensures the load balance amongst slave nodes and distributes and schedules agent model execution tasks to slave modes. It also contains modules of simulation result retrieval and monitoring. Slave nodes execute agent model execution tasks according to the task distribution by the master node. It also reports status information to the master node and synchronizes the execution process with other slave nodes. Experimental results on the parallel platform indicate that it is able to apply various agent based model to the platform with efficient execution cost and good scalability.