HPGA求解流量调和问题的性能评价

针对深水天然气-凝析液生产系统的流量调和问题,采用主从式-粗粒度混合并行遗传算法(Hybrid Parallel Genetic Algorithm,HPGA)求解得到单井流量估计值,以弥补传统遗传算法(Simple Genetic Algorithm,SGA)计算耗时长的缺陷。HPGA基于多核PC集群的分布式储存,通过线程和进程两级并行实现;节点内部使用主从式并行模型(Master-slave Genetic Algorithm,MSGA),在多节点上应用粗粒度并行模型(Coarse-grained Parallel Genetic Algorithm,CGGA)。以某气田两井生产系统为例,通过对比HPGA、MSGA及SGA的计算时间和计算结果,研究HPGA在虚拟计量应用中的并行性能。结果表明:HPGA的并行效率和加速比占线性加速比的比例均在70%以上,计算时间显著缩短,且流量估计误差降低,满足工程运行离线分析的需求。同时,研究了加速比和并行效率随进程数和种群数的变化规律,以探讨并行开销的影响。

Performance evaluation of Hybrid Parallel Genetic Algorithm for solving flow rate reconciliation

To deal with the problem of flow rate reconciliation in deepwater natural gas-condensate production system, master-slave-coarse grained Hybrid Parallel Genetic Algorithm(HPGA) was adopted in this paper to estimate the singlewell flow rate so as to cover the defects of Simple Genetic Algorithm(SGA), i.e., long time consumption for computation. Based on distributed storage of multi-core PC cluster system, HPGA is realized by means of thread and process parallelism. Specifically, Master-Slave Genetic Algorithm(MSGA) is applied within one node, and Coarse-Grained Parallel Genetic Algorithm(CGGA) is adopted among several nodes. Then, case study was carried out on the production systems of two wells in a certain gas field. The parallel performance of HPGA when being applied in virtual flow metering was studied by comparing the computation time and calculation results of HPGA, MSGA and SGA. It is showed that for HPGA, the parallel efficiency and the proportion of speedup ratio to linear speedup ratio are both over 70%. Meanwhile, the computation time is significantly reduced and the error of flow rate estimate is decreased, which satisfies the needs of offline analysis on engineering operation. In addition, the variation of speedup ratio and parallel efficiency with the number of processes and populations was studied so as to discuss the effect of parallel overhead.