A novel information entropy-based decomposition-integration multi-objective evolutionary algorithm for balancing exploration and exploitation

Abstract

Decomposition-based multi-objective evolutionary algorithms (MOEAs) are widely adopted paradigms for solving multi-objective optimization problems (MOPs). While their efficacy is well-established, these algorithms face three core unresolved challenges: inherent difficulty in balancing exploration and exploitation, susceptibility to local optima during neighborhood updates, and Pareto front degradation. Most existing works prioritize objective space optimization, neglecting the non-negligible impact of decision space dispersion on algorithms’ global exploration capability. To address these gaps, this paper proposes an information entropy-based decomposition-integration MOEA (IE-MOEA/DI) to advance the performance of decomposition-based MOEAs. By replacing conventional random initialization with a Riesz s-Energy based uniform population seeding strategy, the algorithm delivers balanced starting points for all subproblems to boost global exploration potential. To address the oversight of decision-space diversity, a new information entropy metric is introduced to quantify population dispersion in the decision space, complementing conventional objective-space-only exploration evaluation. An integrated framework fusing three mainstream scalarization functions is further developed to balance exploration and exploitation, mitigating premature local convergence during rapid neighborhood updates. Complementing these designs, a search angle-guided elite crossover strategy is incorporated to enhance high-quality gene propagation and convergence speed. Extensive experiments on standard benchmark suites verify that IE-MOEA/DI achieves superior overall performance over state-of-the-art decomposition-based MOEAs on IGD⁺ and HV metrics, effectively alleviating Pareto front degradation and realizing a well-balanced exploration-exploitation trade-off.