动态罚函数法求解约束优化问题  

Dynamic Penalty Function Method for Constrained Optimization Problem

作　　者：原杨飞 党乾龙 徐伟 刘玲玲 罗宇婷 YUAN Yangfei;DANG Qianlong;XU Wei;LIU Lingling;LUO Yuting(School of Mathematics and Statistics,Xidian University,Xi’an 710126,China)

机构地区：[1]西安电子科技大学数学与统计学院,西安710126

出　　处：《计算机工程与应用》2022年第4期83-90,共8页Computer Engineering and Applications

基　　金：国家自然科学基金(61772391)。

摘　　要：针对罚函数法在求解约束优化问题时罚系数不易选取的问题,提出一种基于动态罚函数的差分进化算法。利用罚函数法将约束优化问题转化为无约束优化问题。为平衡种群的目标函数和约束违反程度,结合ε约束法设计了一种动态罚系数策略,其中罚系数随着种群质量和进化代数的改变而改变。采用差分进化算法更新种群直到搜索到最优解。对IEEE CEC 2010和IEEE CEC 2017两组基准测试集进行仿真实验,结果表明提出的算法具有较强的寻优性能。Aiming at the disadvantage of the penalty function method that the penalty coefficient is problem-specific parameters,a differential evolution algorithm based on the dynamic penalty function method is presented.First of all,constrained optimization problem is transformed into an unconstrained optimization problem.Afterward,a penalty coefficient adjusting strategy is proposed by combining with theεconstrained method to balance constraint violation and objective function,in which the penalty coefficient changes with the quality of the population and the number of generations.Finally,the differential evolution algorithm is utilized to generate offspring.Experiment results on two benchmark test suites,namely IEEE CEC 2010 and IEEE CEC 2017,demonstrate that the proposed method shows better performance.

关 键 词：约束优化 罚函数法 ε约束法 差分进化 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]