随机性优化算法有效性定量对比评价方法  

Method to compare effectiveness of stochastic optimization algorithm quantitatively and its implementation.

作　　者：李鹏飞[1] 石正军[1] 

机构地区：[1]中国工程物理研究院计算机应用研究所,四川绵阳621900

出　　处：《计算机工程与应用》2014年第13期57-61,共5页Computer Engineering and Applications

基　　金：中国工程物理研究院预研课题

摘　　要：针对随机性优化算法寻优结果不可重复的特点,为该类优化算法提供了一种定量对比评价算法有效性的方法。该方法针对单个或一组测试函数的多次优化结果进行统计分析,得到一个能够在概率意义上定量表征不同随机性算法求解单个或一组测试函数的有效性优劣关系的因子。利用该方法,对采用同步或异步全局最优粒子信息更新模式的两种标准粒子群优化算法(PSO)版本进行有效性对比评价,给出了同步和异步模式PSO算法求解无约束单目标连续变量优化问题的有效性优劣关系。To compare effectiveness of different stochastic optimization algorithms quantitatively, a method is proposed, which is based on statistical analysis of multiple independent results of those algorithms solving a set of typical test samples. Probability quantifying the relative superiority of effectiveness of two algorithms is conducted. Using this method, effec-tiveness comparison of different stochastic optimization algorithms is easy to make. Further, this method is implemented on two patterns of standard Particle Swarm Optimization（PSO）algorithm, in which the global best particle’s information is updated synchronously as well as asynchronously. As a result, a relatively effective pattern of standard PSO algorithm solving unconstrained single-objective optimization problems is advised.

关 键 词：随机性优化算法 有效性评价 粒子群优化算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]