多领域复杂系统模型映射的并行化方法与关键技术研究

This project focuses on the computational efficiency of simulation model for the multi-domain complex engineering systems, and the parallelization method and key technology for CPU-GPU heterogeneous computing resources. Firstly, the dependency and planning optimization strategy of simulation tasks would be analyzed to establish the task graph mapping structure. Then the efficient symbolic and numerical solving strategy for continuous-discrete, linear-nonlinear coupled differential algebraic equations generated from simulation model would be studied to establish solution framework for heterogeneous parallel architectures. Thirdly, the expended parallelism algorithm and speedup technique would be discussed to propose an optimization strategy in the heterogeneous parallelism case for DAE systems. Finally, the validation experiments would be carried out with a complex mechatronics systems named control moment gyro to test and verify some key efficiency characteristics such as the speed-up ratio and computational accuracy.

针对多领域复杂系统仿真模型的计算效率问题，面向CPU与GPU异构计算资源，开展基于仿真模型映射的并行化方法和关键技术研究。在任务并行方面，探索陈述式仿真模型任务的可并行分解及子任务规划方法，形成基于任务图的模型映射结构；在分析求解方面，提出由仿真模型产生的连续和离散混合、线性和非线性耦合的微分代数方程系统(DAE系统)符号与数值综合求解方案，实现适合于异构并行的快速求解框架；在数据并行方面，研究DAE系统自适应异构资源的可扩展并行算法与加速技术,优化DAE系统混合并行条件下的计算策略；最后通过复杂机电产品"控制力矩陀螺"的并行与非并行仿真计算对比，验证仿真模型并行化后的加速比及求解精度等关键效率特性。

针对多领域复杂系统仿真模型的计算效率问题，面向CPU与GPU异构计算资源，开展基于仿真模型映射的并行化方法和关键技术研究。在任务并行方面，探索陈述式仿真模型任务的可并行分解及子任务规划方法，形成基于任务图的模型映射结构；在分析求解方面，提出由仿真模型产生的连续和离散混合、线性和非线性耦合的微分代数方程系统(DAE系统)符号与数值综合求解方案，实现适合于异构并行的快速求解框架；在数据并行方面，研究DAE系统自适应异构资源的可扩展并行算法与加速技术,优化DAE系统混合并行条件下的计算策略；最后通过复杂机电产品"控制力矩陀螺"的并行与非并行仿真计算对比，验证仿真模型并行化后的加速比及求解精度等关键效率特性。