A New Hyper-Heuristic Multi-Objective Optimisation Approach Based on MOEA/D Framework

Area of Science:

• Computational intelligence
• Optimization algorithms
• Evolutionary computation

Background:

• Multi-objective evolutionary algorithms based on decomposition (MOEA/D) are effective for multi-objective optimization problems (MOPs).
• Fixed offspring-generating strategies in MOEA/D can limit applicability, leading to interest in hybrid algorithms.
• Understanding the advantages of hybrid approaches requires investigating dynamic strategy integration.

Purpose of the Study:

• To propose a novel hyper-heuristic approach integrating estimation of distribution (ED) and crossover (CX) strategies into MOEA/D.
• To dynamically switch between differential evolution (DE) and covariance matrix adaptation evolution strategy (CMA-ES) operators.
• To investigate the role of success replacement rate (SRR) in explaining hybrid algorithm advantages.

Main Methods:

• A hyper-heuristic framework is developed within MOEA/D.
• Dynamic switching between DE and CMA-ES operators is implemented.
• Improved Differential Evolution (IDE) is used for specific subproblems to manage evaluation costs.

Main Results:

• The proposed approach demonstrates distinct advantages on a three-objective test suite.
• Significant enhancement in the efficiency (Success Rate Ratio - SRR) of the DE operator is validated.
• Experimental findings provide insights into the performance of hybrid evolutionary algorithms.

Conclusions:

• The novel hyper-heuristic MOEA/D approach effectively integrates ED and CX strategies.
• Dynamic operator switching enhances performance and efficiency in multi-objective optimization.
• The study offers valuable perspectives on hybrid algorithms and SRR for future research.