A newly proposed evolutionary computation framework, designated as MOEA/D-DWALA, demonstrates competitive or superior performance in convergence, diversity, and efficiency across multi-objective optimization tasks. The technique, which introduces an entropy-guided decomposition-based multi-objective evolutionary algorithm with dynamic weight adjustment and local acceleration, was formulated to address persistent algorithmic challenges encountered when optimizing complex problem landscapes with irregular Pareto fronts.
In decomposition-based multi-objective evolutionary algorithms, known commonly as MOEA/D, the performance of the search process depends strongly upon the distribution of its weight vectors. When optimization routines rely on fixed or insufficiently adapted weight vectors, those vectors frequently mismatch irregular Pareto fronts. This geometric mismatch causes significant operational degradation, leading directly to inactive subproblems, overly crowded search directions, and degraded diversity across candidate solutions.
To address this issue directly, MOEA/D-DWALA integrates an entropy-guided archive designed to identify and preserve representative elite solutions that exhibit low redundancy. Furthermore, to counteract the high computational costs typically associated with frequent archive maintenance routines, the algorithm introduces local subset selection. Alongside archive management, the system continuously updates weight vectors according to the estimated distribution of the Pareto front.
These interconnected algorithmic components function collectively as a closed-loop mechanism during execution. Within this feedback loop, the entropy-filtered archive provides continuous informational feedback to drive dynamic weight redistribution across the objective space. Simultaneously, the incorporated local acceleration strategy works to lower computational overhead and enhance overall computational efficiency throughout the iterative evolutionary search process.
The performance of the proposed MOEA/D-DWALA algorithm was systematically evaluated across experimental trials encompassing both benchmark test problems and real-world many-objective engineering problems. The resulting experimental findings confirmed that the closed-loop algorithm consistently achieves competitive or superior performance across convergence, solution diversity, and computational efficiency when evaluated on these challenging optimization tasks.
Subject of Research: Multi-objective evolutionary algorithms and dynamic weight adjustment
Article Title: Entropy-guided decomposition-based multi-objective evolutionary algorithm with dynamic weight adjustment and local acceleration
Article References: Ling, Q., Wang, Z., Ji, Z., Shi, J., & Wang, L. (2027). Entropy-guided decomposition-based multi-objective evolutionary algorithm with dynamic weight adjustment and local acceleration. Cluster Computing, 30(1), Article 4. https://doi.org/10.1007/s10586-026-06795-2
Image Credits: AI Generated
DOI: 10.1007/s10586-026-06795-2
Keywords: evolutionary algorithms, multi-objective optimization, decomposition-based algorithms, MOEA/D, MOEA/D-DWALA, entropy-guided archive, dynamic weight adjustment, local acceleration, Pareto front, solution diversity, algorithm convergence, computational efficiency, local subset selection, many-objective engineering, closed-loop optimization, subproblem optimization, weight vector distribution
News Source: Gavin Prescott. (November 16, 2026). Closed-Loop Evolutionary Algorithm Improves Convergence and Diversity in Multi-Objective Optimization. Scienmag.



