基于权重迭代的偏好多目标分解算法解决参考点对算法影响的研究

在传统偏好多目标进化算法中,参考点是表达决策者的偏好信息最常用的方式,但是参考点所处位置信息有时严重影响算法的性能.针对以上问题,本文提出了一种基于权重迭代的偏好多目标分解算法(MOEA/D-PRE),主要利用权重迭代方法获取一组均匀的权重向量,并对偏好区域进行映射,使得算法在进化过程中,不用考虑参考点所处位置信息对算法性能的影响,另外提出了一种稳定可控的偏好区域模型,能响应决策者设置任意大小的偏好区域.通过对比实验表明该算法具有较好的收敛性和分布性,同时给出了满足决策者不同要求的算法模型,并且能够很好的解决参考点的位置信息对算法的影响.

Research on MOEA/D Based on User-Preference and Alternate Weight to Solve the Effect of Reference Point on Multi-Objective Algorithms

In MOEAs based on user-preference, reference point is most commonly used to express the preference information, but the position of reference point has detrimental effect on the performance of algorithms.According to this issue, this paper proposes MOEA/D-PRE that combines MOEA/D with preference information and alternate weight method.This algorithm applies the alternate weight method to map the region of interest of the decision maker, which can avoid the influence of the reference point, and the model is easy for the decision maker to adjust the size of preference region.Experimental results show that this approach has much better performance.Moreover, this paper proposes different models to satisfy different demands of the decision maker, which has provided a new way to solve MOPs based on preference information and especially to tackle the effect of reference point.