基于CUDA的并行粒子群优化算法研究及实现

Research and Design of Parallel Particle Swarm Optimization Algorithm Based on CUDA

应用图形处理器(GPU)来加速粒子群优化(PSO)算法并行计算时,为突出其加速性能,经常有文献以恶化CPU端PSO算法性能为代价。为了科学比较GPU-PSO算法和CPU-PSO算法的性能,提出用"有效加速比"作为算法的性能指标。文中给出的评价方法不需要CPU和GPU端粒子数相同,将GPU并行算法与最优CPU串行算法的性能作比较,以加速收敛到目标精度为准则,在统一计算设备架构(CUDA)下对多个基准测试函数进行了数值仿真实验。结果表明,在GPU上大幅增加粒子数能够加速PSO算法收敛到目标精度,与CPU-PSO相比,获得了10倍以上的"有效加速比"。

In the application of graphic processing unit （GPU） to accelerate particle swarm optimization （PSO） algorithm for parallel computing,many references worsen the performance of PSO algorithm on CPU side in order to highlight the acceleration performance. The concept of ＂Effective Speedup＂ was proposed in this paper to measure the achievement of GPU-PSO algorithm and CPU-PSO algorithm. The proposed method aims at accelerating the implementation to the target precision. The GPU parallel algorithm was compared with the best CPU serial algorithm, which does not require the same number of particles between CPU side and GPU side. Experiments based on several benchmark test functions using compute unified device architecture （CUDA） show that substantially increasing the number of particles on GPU side can significantly accelerate the accomplishment of PSO algorithm to the target precision. Compared with CPU-PSO, an ＂Effective Speedup＂ of more than 10 has been achieved.