摘要
针对处理大量数据和求解大规模复杂问题时粒子群优化(PSO)算法计算时间过长的问题,进行了在显卡(GPU)上实现细粒度并行粒子群算法的研究。通过对传统PSO算法的分析,结合目前被广泛使用的基于GPU的并行计算技术,设计实现了一种并行PSO方法。本方法的执行基于统一计算架构(CUDA),使用大量的GPU线程并行处理各个粒子的搜索过程来加速整个粒子群的收敛速度。程序充分使用CUDA自带的各种数学计算库,从而保证了程序的稳定性和易写性。通过对多个基准优化测试函数的求解证明,相对于基于CPU的串行计算方法,在求解收敛性一致的前提下,基于CUDA架构的并行PSO求解方法可以取得高达90倍的计算加速比。
This paper raised a fine-grained PSO algorism based on GPU acceleration, which could reduce the computing time for processing large amounts of data and solve large-scale complex problems. The implementation of proposed method based on compute unified device architecture （ CUDA）, in order to accelerate the convergence rate of whole swarm, a larger number of GPU threads used to parallel process a single update and fitness evaluation alone. For ensuring the stability of the code and it easier to program, fully used several numerical library provide by CUDA. Experiments based on several benchmark test func- tions show that more than 90 times speeds obtained with the same calculation precision, it compared to CPU-based sequential implementation.
出处
《计算机应用研究》
CSCD
北大核心
2013年第8期2415-2418,共4页
Application Research of Computers
基金
国家"863"计划资助项目(2012AA111802)
国家自然科学基金资助项目(11172097)
