基于角度惩罚距离精英选择策略的偏好高维目标优化算法

发布时间：2018-06-09 05:35

本文选题：高维目标优化 + 偏好向量　；参考：《计算机学报》2018年01期

【摘要】：基于决策者偏好的高维目标优化算法能有效集中算法资源和减小搜索空间,是处理高维目标优化问题的有效途径之一.现有研究发现,参考点位置选择对算法性能影响显著,位于极端位置的参考点容易引发算法不收敛;同时,算法多样性在种群逼近Pareto前沿的过程中反复遭到破坏.为解决以上问题,该文提出一种基于角度惩罚距离精英选择策略的偏好高维目标优化算法.该算法将决策者偏好信息融入到基于分解的多目标优化算法中,提出偏好向量生成策略,消除算法收敛性对参考点位置的敏感性;同时引入角度惩罚距离(APD)机制,分析该机制在算法搜索后期存在种群退化、收敛放缓等缺陷的基础上,提出APD精英选择策略,通过有效分配算法资源,平衡算法收敛性和多样性.算法性能对比实验中,将该文提出的算法与g-占优、r-占优、双极偏好占优以及MOEA/D-PRE在3至10维DTLZ1-4测试问题上进行性能测试.实验结果表明,该文提出的偏好算法所求解集能够有效反映决策者的偏好信息,并且在高维目标优化问题上,所提算法在偏好区域求得解集的收敛性和均匀性更优.
[Abstract]:The high dimensional objective optimization algorithm based on the decision maker's preference can effectively concentrate the resource of the algorithm and reduce the search space. It is one of the effective ways to deal with the optimization of the high dimension target. The existing research finds that the location of the reference points has a significant influence on the performance of the algorithm, and the reference points in the extreme position can easily lead to the algorithm's non convergence; at the same time, the algorithm is diverse. In order to solve the above problem, this paper proposes a preference high dimension target optimization algorithm based on the angle penalty distance elite selection strategy. This algorithm integrates the decision maker's preference information into the decomposition based multi-objective optimization algorithm, and proposes a preference vector generation strategy to eliminate the calculation. The sensitivity of the law of convergence to the location of the reference point; at the same time introducing the angle penalty distance (APD) mechanism, the mechanism is analyzed on the basis of the defects such as the population degradation and the slowing down of convergence in the later period of the algorithm search, and the APD elite selection strategy is proposed to balance the convergence and diversity of the algorithm by effectively distributing the algorithm resources. The algorithm proposed in this paper is dominated by g-, r- dominance, bipolar preference and MOEA/D-PRE performance testing on 3 to 10 dimensional DTLZ1-4 test problems. Experimental results show that the set of preference algorithms proposed in this paper can effectively reflect the preference information of the decision-makers, and the proposed algorithm is in the preference region for the high dimensional target optimization problem. The convergence and uniformity of the obtained solution set are better.
【作者单位】：浙江工业大学信息智能与决策优化研究所;浙江工业大学教育科学与技术学院;浙江工业大学现代教育技术研究所;
【基金】：国家自然科学基金(61472366,61379077,61503340) 浙江省自然科学基金(LY17F020022,LQ16F030008)资助
【分类号】：TP18

【相似文献】