摘要
并行仿真是提高仿真性能的关键技术,然而当前基于FMI(Functional Mock-up Interface)的并行联合仿真存在诸多挑战,如FMU(Functional Mock-up Unit)和执行线程耦合、输入输出同步、FMI API互斥等问题。针对这些问题,提出了一种基于线程池任务调度的局部同步FMI联合仿真方法。首先,给出了该方法的框架,该框架由仿真方案、主算法、调度器和缓冲区组成,以提供并行仿真的模块化表示。然后,重点描述了FMI并行联合仿真主算法,其将单个FMU的单次仿真任务分为仿真执行和任务调度两阶段,由调度器调度执行,并自定义读写锁解决仿真过程中的同步问题;同时,通过将输出暂存到缓冲区,以避免FMI API竞争访问的问题。最后通过一个房间温差模型和一个船舶定位模型验证了所提方法的准确性,与FMU并行的非迭代雅可比方法进行对比,所提方法取得了显著的性能提升。
Parallel simulation is a key means to improve simulation performance.However,parallel co-simulation based on FMI faces many challenges,such as coupling between FMU and threads,input/output synchronization,and mutual exclusion of FMI API.In response to these problems,this paper proposes a local synchronous FMI co-simulation method based on thread pool task scheduling.Firstly,the framework of the method is presented,consisting of a simulation scheme,master algorithm,scheduler,and buffer to provide a modular representation of parallel simulation.Then,the master algorithm of FMI parallel co-simulation is described,which divides a simulation task of a FMU into two stages:simulation execution and task scheduling.The scheduler sche-dules the task to execute.And customizes read-write lock is designed to solve the synchronization problem during the simulation.The output is temporarily stored in a buffer to solve the problem of FMI API contention for access.The accuracy of the proposed method is verified through a room temperature difference model and a ship positioning model.Compared with the non-iterative Jacobi method parallel to FMU,significant performance improvement is achieved.
作者
薛朝阳
钱晓超
刘飞
XUE Zhaoyang;QIAN Xiaochao;LIU Fei(School of Software Engineering,South China University of Technology,Guangzhou 515000,China;Shanghai Institute of Mechanical and Electrical Engineering,Shanghai 201109,China)
出处
《计算机科学》
北大核心
2026年第2期107-116,共10页
Computer Science
基金
国家自然科学基金(62273153)
广东省基础与应用基础研究基金(2024A1515010900)。
关键词
并行仿真
联合仿真
FMI
FMU
主算法
无锁优化
Parallel simulation
Co-simulation
Functional Mock-up Interface
Functional Mock-up Unit
Master algorithm
Lock-free
