SPAM: Set Preference Algorithm for Multiobjective Optimization
SPAM: Set Preference Algorithm for Multiobjective Optimization
复制标题
DOI：
10.1007/978-3-540-87700-4_84
复制
发表时间：
2008-09
期刊：
影响因子：
--
通讯作者：
E. Zitzler;L. Thiele;J. Bader
中科院分区：
文献类型：
--
作者：
E. Zitzler;L. Thiele;J. Bader
本文追求的是通用多目标优化器的思想，它可以灵活地适应任意用户偏好——假设目标是逼近帕累托最优集。它提出了多目标优化（SPAM）的集合偏好算法，其工作原理基于两个观察：（i）当前的多目标进化算法（MOEA）可以被视为集合问题上的登山者，（ii）特定的用户偏好通常（隐式）以帕累托集合近似上的二元关系来表达。 SPAM 在 Pareto 集合近似空间上实现 (1 + 1) 策略，并且可以与任何类型的集合偏好关系一起使用，即定义 Pareto 集合近似上的总先序的二元关系。实验结果表明，对于一系列设定的偏好关系，SPAM 为用户偏好提供了充分的灵活性，并且可以根据指定的偏好进行有效的优化。因此，它为偏好引导的多目标搜索提供了新的视角。
This paper pursues the idea of a general multiobjective optimizer that can be flexibly adapted to arbitrary user preferences—assuming that the goal is to approximate the Pareto-optimal set. It proposes the Set Preference Algorithm for Multiobjective Optimization (SPAM) the working principle of which is based on two observations: (i) current multiobjective evolutionary algorithms (MOEAs) can be regarded as hill climbers on set problems and (ii) specific user preferences are often (implicitly) expressed in terms of a binary relation on Pareto set approximations. SPAM realizes a (1 + 1)-strategy on the space of Pareto set approximations and can be used with any type of set preference relations, i.e., binary relations that define a total preorder on Pareto set approximations. The experimental results demonstrate for a range of set preference relations that SPAM provides full flexibility with respect to user preferences and is effective in optimizing according to the specified preferences. It thereby offers a new perspective on preference-guided multiobjective search.