基于多偏好自适应协同的高维目标进化算法  

Many-objective Optimization Via Cooperative Coevolution with Adaptive Preferences

作　　者：王丽萍[1,2] 彭骏[1,2] 邱飞岳[3] 莫雷平 

机构地区：[1]浙江工业大学经贸学院,杭州310023 [2]浙江工业大学智能信息处理研究所,杭州310023 [3]浙江工业大学教育科学与技术学院,杭州310023 [4]浙江工业大学信息工程学院,杭州310023

出　　处：《小型微型计算机系统》2016年第6期1308-1312,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61472366;61379077)资助;浙江省自然科学基金项目(LZ13F020002;LY13F030010)资助

摘　　要：在解决高维目标优化问题时,针对多偏好的优化解决方案存在偏好值设置困难和计算复杂度高的缺点,提出一种基于多偏好自适应协同进化的高维目标进化算法.首先利用多个随机偏好引导种群进化,然后利用进化后的种群选择偏好,这两个过程循环交替,实现偏好与种群协同进化来.在协同进化的基础上,为了在最大化偏好的效用性的同时降低计算复杂,进一步提出多偏好自适应策略,即在混合排序的机制下,给与种群一个可变的偏好适应周期.在算法性能仿真实验中,该算法应用于求解2到10目标的WFG测试函数,评估了算法计算复杂度和超体积指标,实验结果表明,在保持解集质量的同时能够有效降低算法的计算复杂度,从而有效改善了解集质量与算法复杂度之间的平衡.The main difficulties of solving many-objective optimization problems using preferences-based evolutionary algorithms are preferences values initialization and high computational complexity. In this work, a new many-objective optimization algorithm via co- operative coevolution with adaptive preferences was proposed for solving optimization problems with more than four objectives. Firstly the randomly generated preferences were introduced to guide the population and the evoluated population selects preferences. These processes run iteratively to realize cooperative coevolution between preferences and population. The concept of coevolving a family of preferences together with population was used to improve efficiency of preferences of update and feedback. Secondly, an adaptive pref- erences strategy based on cooperative coevolution was proposed to improve the effectivity of preferences and reduce the computing complexity. In the experimemal analysis, the proposed algorithm was tested on the WFG problems with two-, seven-and ten-objectives. The results demonstrated that this approach can provide solutions with high-quality and reduce the computational complexity when sol- ving many-objective problems.

关 键 词：高维多目标 多偏好 自适应 协同进化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]