一种基于有限K近邻的强度帕累托进化算法  

An Improved Strength Pareto Evolutionary Algorithm Based on the Limited K-Nearest Neighbor Method

作　　者：姜宏[1,2] 杨孟飞[3,2] 于广良[1,2] 魏梦捷 

机构地区：[1]北京控制工程研究所,北京100190 [2]空间智能控制技术重点实验室,北京100190 [3]中国空间技术研究院,北京100094

出　　处：《中国空间科学技术》2015年第2期49-56,共8页Chinese Space Science and Technology

基　　金：国家自然科学基金(91118007)资助项目

摘　　要：在航天器控制计算机的软硬件协同设计过程中,需要解决多目标优化问题。当前的强度帕累托进化算法在求解高维多目标优化问题时具有优势,但是在环境选择阶段的计算时间复杂度仍然较大。文章针对这一问题,提出了一种改进算法。新的算法采用有限K近邻方法,减少了原算法中K近邻策略的比较次数,使时间复杂度由O(M3)下降为O(max(l,log M)M2)。试验结果表明文中算法的计算速度更快,并且具有更优的收敛性和分布多样性特征。In the process of Hardware/software co-design of spacecraft control computers, the multi-objective optimization is a key problem. The current strength Pareto evoIutionary algorithm has some advantages in solving high-dimensional multi-objective optimization problems, but the computing time complexity during the step of environmental selection is still very large. Aiming at this point, an improved algorithm was proposed. With the finite K-nearest neighbor method, new algorithm reduces the number of comparisons to lower the time-complexity from O（Ma） down to O（max（l, logM）M2）. The experimental results show that the proposed algorithm not only improves the running speed, but also acquires better convergence and distribution diversity than the original one.

关 键 词：软硬件协同设计 多目标优化 帕累托最优 强度帕累托进化算法 星载计算机 航天器控制 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]