Analysis of Multi-objective Hyper-Heuristics Under Different Dynamic and Preferential Environments

Abstract

The use of hyper-heuristics to solve dynamic multi-objective optimization problems (DMOPs) that incorporate decision-maker's preferences is a recently addressed research area. This paper proposes the analysis and comparison of three hyper-heuristics to solve preferential DMOPs. The Dynamic Hyper-Heuristic with Plane Separation (DHH-PS), a previously proposed methodology using Plane Separation (PS), a reference-point-based preference incorporation method. This paper also proposes two versions of the Dynamic Population-Evolvability based Multi-objective Hyper-Heuristic (DPEM-HH), incorporating PS and different low-level heuristics sets. This work tests DHH-PS and both DPEM-HH-PS versions under multiple dynamic and preferential environments, seeking to extend the study of DHH-PS and analyze the capability of DPEM-HH-PS. DPEM-HH-PS exhibited suitability for type II DMOPs and randomly-changing instances. DHH-PS presented a better performance for tri-objective DMOPs. The combination of genetic algorithms and differential evolution in DPEM-HH-PS proved effective for solving preferential DMOPs. DHH-PS and DPEM-HH-PS were capable of adapting to different preferential and dynamic environments.