一种基于MPI的河流生态修复模拟并行算法

A MPI-based parallel computing method for the simulation of river ecological restoration

河流生态修复模拟评估过程中修复情景的大批量产生特性以及模拟计算结果的高精度要求,使得海量情景的模拟计算工作量巨大。为提高情景模拟计算速度和缩短决策时间,将并行计算技术引入河流生态修复模拟评估过程,设计了基于MPI的并行计算算法,实现了模拟计算的并行化处理。算法包括读取决策变量取值集合和约束条件集合、决策变量取值集合分解、情景生成、情景模拟、模拟结果写入文件等主要步骤,其中设计了自适应集合分解子算法,用以解决并行处理中的负载均衡问题。以永定河生态修复为例,在单机多核计算机平台上应用算法进行了对比实验,在4核平台上使模拟计算时间减少为串行时的近四分之一,加速比达3.8,在万级、十万级和百万级情景数量规模下均保持稳定一致的加速比。实验结果表明:该算法性能稳定,实用性强,可有效缩短海量情景模拟的计算时间,显著提高计算效率。

Large quantities of scenarios could be generated in the decision-making process of river ecosystem restoration,and thus led to huge computing workloads of simulation and assessment,especially when a high-precision of the results is needed.A parallel computing method based on the Message-Passing Interface（MPI）standard is proposed in the paper to shorten the calculation time and meet the efficiency requirement.The implementation procedure of this method includes reading the variables and constraints from the input files,dividing the variables into sub-sets for different processes,generating scenarios,simulating the results of scenarios,and finally writing the results into the output files.In particular,an adaptive set dividing algorithm was developed to balance the workloads of all the processes in the parallel computing.By taking the ecological restoration of the Yongding River as a case study,the proposed method was applied in a decision support system and tested on two computers with different multi-core CPUs.The experimental results obtained showed that when performed on a computer with a 4-core CPU,the method reduced the calculation time to a quarter of the normal computation time,with a stable speedup factor of 3.8at different amounts of scenarios.It is further indicated that,generally speaking,the proposed method is practical and it can effectively shorten the execution time of simulation of massive scenarios and significantly improve the computational efficiency with a stable performance.