摘要
Evolutionary algorithms have shown prominent performance in solving various kinds of multi-objective optimization problems (MOPs), but most of them only use single operator that is often sensitive to the characteristics of problems. As different operators have different search patterns, a proper combination of multiple operators can be more efficient and robust than using one single operator in solving complex problems. However, how to ensemble multiple operators based on their performances in optimization process is a challenging task. In the machine learning field, it is well known that AdaBoost can effectively ensemble multiple classifiers by giving the proper weights based on their classification errors. Inspired by this ensemble way, we propose a multi-operator ensemble (MOE) strategy based on multiple subpopulations for evolutionary multi-objective optimization. In the proposed strategy, the survival rate of each subpopulation after environmental selection is used to evaluate the performance of the operator, and then a credit assignment scheme is developed by using the weight update method in AdaBoost. Based on these credits, an emigration-immigration mechanism is designed to update the subpopulation that can adaptively reward or punish the computational resources for operators. Experimental results on three complex test suites demonstrate that the proposed MOE can significantly improve the performance of multi-objective evolutionary algorithms on different types of MOPs.
学科与关键词
- An effective ensemble framework for Many-Objective optimization based on AdaBoost and K-means clusteringPalakonda, Vikas; Kang, Jae-Mo; Jung, Heechul*EXPERT SYSTEMS WITH APPLICATIONS, 2023, 227: 120278.被引 7
- A Classification and Comparison of Credit Assignment Strategies in Multiobjective Adaptive Operator SelectionHitomi, Nozomi*; Selva, DanielIEEE Transactions on Evolutionary Computation, 2017, 21(2): 294-314.被引 30
- A Multioperator Search Strategy Based on Cheap Surrogate Models for Evolutionary OptimizationGong Wenyin; Zhou Aimin*; Cai ZhihuaIEEE Transactions on Evolutionary Computation, 2015, 19(5): 746-758.
