不同拓扑结构的并行粒子群优化算法的实现

摘要

针对粒子群优化算法的邻域拓扑结构对算法性能有重要影响、PSO 算法在 CPU 上求解最优化问题时计算效率低下这两点，分析了邻域拓扑结构改变时 PSO 算法的并行特征，实现了环形和星形拓扑结构的PSO算法在统一计算设备架构上的寻优过程。分别在CPU和GPU上用两种PSO算法对7个benchmark测试函数进行求解。程序仿真结果显示，基于CUDA的PSO算法计算效率均大大高于CPU；同时发现，GPU显著地加快了星形结构PSO算法的收敛速度，而对环形结构PSO算法影响不大。%Neighborhood topology has an important influence on the performance of particle swarm optimization algorithm. The algorithm for solving optimization problems on the CPU is very inefficient. For these two point, analyzing parallel characteristic of PSO algorithm when neighborhood topology changes and achieving a ring and star topologies PSO algorithm on compute unified device architecture(CUDA) on the optimization process. Solving 7 benchmark test functions on the CPU and the GPU PSO algorithm respectively, the program simulation results show that PSO algorithm based on CUDA computing efficiency is significantly higher than CPU. In the meantime, GPU accelerates dramatically star PSO algorithm convergence speed, while the ring structure PSO algorithm have little effect.