一种环境理解的分布式群体仿真任务划分方法  

An Environment Structure Understanding Task Partition Method for Distributed Crowd Simulation

作　　者：周文平[1] 唐好选[1] 季振洲[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,黑龙江哈尔滨150001

出　　处：《电子学报》2014年第12期2448-2456,共9页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61175027)

摘　　要：本文通过引入环境结构因素,提出了一种适用于多层次复杂环境的自适应任务划分算法.自动读取场景模型并通过理解转换为连通邻接区域集,然后对区域进行快速粗粒度划分,有效提高划分性能;自然消除了被障碍隔离的相邻区域个体间的感知计算,大大减少了节点间通信量,使之更适合于大规模群体仿真应用.实验结果表明该算法的划分代价和执行性能均较优.文中设计了一种适合该划分算法的分布式仿真模型,基于该模型的分布式系统对室内多层楼宇或室外场景大规模群体仿真均具有较高仿真性能,相同规模群体的仿真性能与仿真节点数成线性关系表明系统具有良好的可扩展性.The paper proposed an adaptive partition method considering environmental structural factors,which is applicable for multilayered complex environment .Simulation scene is automatically extracted to walkable adjacent areas,and a coarse granular-ity partition based on regions is applied to get shorter execution time .The inter-individual perceptual computing of any two individu-als separated by obstacle between two adjacent regions is negligible,so it efficiently reduces the inter-node communication cost and makes the algorithm more suitable for large scale crowd simulation .The results show the proposed algorithm gets lower cost and higher performance .An efficient distributed simulation model is designed for the partition method,and a distributed system based on the model gets higher simulation performance on both inner door and out door scene .The performance of system with the same crowd size linearly increases with the increase of compute nodes,which proves high scalability of the system .

关 键 词：分布式仿真 群体仿真 任务划分 图形处理单元 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]