Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
6864616 | Neurocomputing | 2018 | 16 Pages |
Abstract
It is challenging to solve reducible many-objective problems due to difficulties caused by the unknown number of non-conflicting objectives. Objective reduction method is one of promising and efficient solutions in which two fundamental problems should be addressed: how to find the redundant objectives and which objectives should be selected or omitted. A novel objective reduction algorithm is proposed in this paper, named Maximal Information Coefficient based Multi-Objective Particle Swarm Optimizer (MIC-MOPSO). By a powerful MIC indicator, the algorithm could find hidden linear or nonlinear relationships between two objectives. Another indicator, the change rate of non-dominated population, is used to judge whether there exist non-conflicting objectives or not. An effective way to rapidly select the retained objectives is also developed based on these two indicators. Tested by a series of benchmark experiments and a real industrial optimization problem, the results show that our approach significantly improve the performance on both reducible and irreducible many-objective problems.
Related Topics
Physical Sciences and Engineering
Computer Science
Artificial Intelligence
Authors
Liang Yi, He Wangli, Zhong Weimin, Qian Feng,