基于粒子群优化与高斯过程的协同优化算法  

Cooperative optimization algorithm based on particle swarm optimization and Gaussian process

作　　者：张研[1,2] 苏国韶[1,3] 燕柳斌[1] 

机构地区：[1]广西大学土木建筑工程学院,广西南宁530004 [2]广西大学工程防灾与结构安全教育部重点实验室,广西南宁530004 [3]广西防灾减灾与工程安全重点实验室,广西南宁530004

出　　处：《系统工程与电子技术》2013年第6期1342-1347,共6页Systems Engineering and Electronics

基　　金：国家自然科学基金(51069001);广西理工科学实验中心重点项目(LGZX201001);广西重点实验室系统性研究项目(2012ZDX10)资助课题

摘　　要：对于适应度函数计算耗时较大的工程优化问题,采用仿生智能优化算法求解时常遇到由于适应度函数评价次数过大而导致计算量过高的瓶颈问题。针对上述问题,提出一种基于粒子群优化(particle swarm opti-mization,PSO)算法与高斯过程(Gaussian process,GP)机器学习方法的协同优化算法(PSO-GP)。该算法在寻优过程中采用GP近似模型来构建决策变量与适应度函数值之间的映射关系,在PSO全局寻优过程中不断地总结寻优历史经验的基础上,预测可能包含全局最优解的搜索区域,以优化粒子群飞行的方向。多个测试函数的优化结果表明,该算法是可行的,与基本PSO算法相比,在获得全局最优解的前提下,可显著减小寻优过程中的适应度函数评价次数,寻优效率较高,在高计算代价复杂工程优化问题的求解上具有良好的应用前景。The large numbers of fitness function evaluation are needed when the engineering optimization problems with time consuming fitness evaluations are solved using a bionic intelligent optimization algorithm. This poses a serious impediment to the field of the bionic intelligent optimization algorithm for the unacceptable high cost of calculation. A cooperative optimization algorithm based on particle swarm optimization （PSO） algo- rithm and Gaussian process （GP） machine learning for solving computationally expensive optimization problems is presented. GP is used as a surrogate of the real fitness function to prevent frequent fitness function evaluation and predict the most promising solutions before searching the global optimum solution using PSO during each iteration step. The results of study show that the proposed algorithm is much more economical to achieve reasonable accuracy with much less fitness evaluations when solving the optimization problems of the benchmark functions compared with the basic PSO. The proposed algorithm seems very promising to solve the time-consuming optimization problems.

关 键 词：优化算法 粒子群优化 高斯过程 函数优化 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]