摘要
基于统一计算设备架构构建的图形处理器(Graphics Processing Unit,GPU)众多运算颗粒,可用于遗传算法个体业务函数的并行调用以起到加速遗传算法的作用。本文重点研究GPU加速遗传算法的程序实现,并针对低配置GPU不支持双精度浮点运算的不足,通过扩展字节等方式构造出双浮点精度的改良算法。随后将改进后的遗传算法用于研究Halo轨道的生成问题,克服传统算法对6×6维状态转移矩阵等先决条件的需求。优化结果表明GPU加速性能和双浮点精度改进设计等效果良好。
This paper deals with the application of the Computing Unified Device Architecture （CUDA） into the constructing methodology for Halo orbit, which employs many parallel kernels to accelerate the individual professional functions of genetic algorithm. It focuses on the programming implementation of （Graphics Processing Unit, GPU） accelerating genetic algorithm, and settles with the CUDA＇s trouble of single-precision floating-point operation by the way of expanding the double-byte floating-point precision. And then the improved genetic algorithm in constructing Halo orbit has overcome the traditional deficiency which requires the 6×6-dimensional state transition matrix. Optimization results show that GPU acceleration to improve performance and double-precision floating-point design to good effect.
出处
《科研信息化技术与应用》
2011年第6期104-112,共9页
E-science Technology & Application
基金
国家自然科学基金(11172020)
中国航天科技集团公司航天科技创新基金资助项目
工信部"唯实"人才培育基金(YWF-11-03-Q-064)
北京航空航天大学"蓝天新秀"专项基金
