【摘要】 目的针对传统遗传算法容易陷于极值,计算时间长的问题,设计基于计算机集群的一种新的粗粒度并行遗传算法反演场地浅层剪切波速度结构.方法采用遗传模拟退火算法和MPI并行计算技术,实现多进程的粗粒度集群计算,通过个体迁移策略协调优化子种群,运用计算效率判断计算负载状态,采用动态种群进行负载平衡,构建了4节点的PC集群,对算例和实际场地的浅层剪切波速度结构进行了反演计算.结果简单模型收敛于最优解,实际场地的反演结果与钻孔资料的平均误差均在20%以内,计算速度明显提高,并行遗传算法的反演结果好于串行遗传算法反演的结果.结论笔者设计的粗粒度并行遗传算法有效地加快了进化速度,并行效率高,加强了局部搜索能力,反演结果较好,适合应用于反演实际工程场地的浅层剪切波速度结构.更多还原

【Abstract】 A coarse-grained parallel genetic algorithm based on personal computer cluster is proposed to inverse shallow S-wave velocity structure of actual engineering sites,in order to solve the disadvantage of traditional genetic algorithm,such as easy trap into extreme and computed long time.The simulated annealing algorithm and parallel technique message passing interface(MPI)are adopted to implement the coarse-grained parallel computer.The subpopulations are collaboratively optimized through individual migration strategy and the dynamic populations are adopted to balance the computing load.The shallow shear wave velocity structures of two examples and the actual engineering sites are inversed through a 4-node PC cluster test system.The simple models converge to the optimum solution and the average errors of the inversed results of the actual engineering sites are all less than 20% compared with the PS logging data.The computing speed increases rapidly,and the results inversed by parallel genetic algorithm are better than those inversed by genetic algorithm.The coarse-grained parallel genetic algorithm designed increases the evolutionary rate effectively and the algorithm has a good parallel efficiency.The locally search capability in inversion is improved and the inversed results are better.The results show that the algorithm proposed can be used to inverse shallow S-wave velocity structure of engineering site.更多还原