A constrained multi-objective evolutionary algorithm with clustering based weight vector adaptation

Sri Srinivasa Raju M

doi:10.1016/j.swevo.2023.101432

A constrained multi-objective evolutionary algorithm with clustering based weight vector adaptation

2024-02-01

1

PRE

AI

Rammohan Mallipeddi *

DOI：10.1016/j.swevo.2023.101432

Original

Original request for help

Save

Abstract

En 中文

In constrained multi-objective optimization, balancing convergence, diversity, and feasibility of solutions in the population are challenging. Most existing approaches focus on convergence and feasibility but fail to enforce the diversity and uniformity of solutions in the approximated Pareto Front (PF). In decomposition-based approaches, the diversity and uniformity of the obtained solution set depend on the set of weight vectors employed, which are generally fixed and initialized uniformly. However, Constrained Multi-objective Optimization Problems (CMOPs) are characterized by discontinuous Pareto-optimal fronts due to the presence of constraints. Therefore, to maintain better diversity and uniformity in the population, the weight vector set employed in the framework should consider the characteristics of PF, which is not known in advance. However, during the evolution, the characteristics of PF can be estimated and the associated weight vectors can be appropriately distributed. Motivated by this issue, we propose a novel Constrained Multi-Objective Evolutionary Algorithm with a Clustering-based Weight vector Adaptation strategy called CMOEA-CWA. The proposed CMOEA-CWA uses the basic framework of CMOEA-DPMS. After exploring the search space, CMOEA-CWA reinitializes the weight vectors if there exist any discontinuities in the approximated pareto front (PF). In CMOEA-CWA, the PF is approximated using the feasible solutions stored in an archive and novel parameter-free clustering techniques. During the clustering process, if there exist any discontinuities in the approximated PF, one population will exploit the feasible regions with a re-initialized set of weight vectors while the other population will explore the feasible regions with uniformly distributed weight vectors. Empirical results obtained on four popular and widely used benchmarks when compared with nine latest state-of-the-art algorithms (ARMOEA, CCMO, TiGE-2, ToP, MOEA/D-DAE, PPS, CTAEA, MOEA/D-2WA and CMOEA-DPMS) suggest the superiority of the proposed algo-rithm, especially on 47 complicated constrained multi-objective problems. The proposed algorithm performs statistically better than or comparable (in terms of HV) to ARMOEA, CCMO, TiGE-2, ToP, MOEA/D-DAE, PPS, CTAEA, MOEA/D-2WA and CMOEA-DPMS in 91.48 %, 78.72 %, 100 %, 70.21 %, 85.11 %, 78.72 %, 91.49 % and 82.97 % of test instances, respectively. The performance of the proposed algorithm is also demonstrated on a real-world Economic Environmental Dispatch (EED) problem.

Keywords:

Weight vector adaptation

Clustering

Constraint multi -objective evolutionary algo

rithm

Optimization

Multi -staged

Dual population

A constrained multi-objective evolutionary algorithm with clustering based weight vector adaptation

Abstract

Journal

Organization

Cited Papers