Reinforcement learning assisted automatic niche selection for constrained multimodal multi-objective optimization

doi:10.1016/j.eswa.2025.129458

Reinforcement learning assisted automatic niche selection for constrained multimodal multi-objective optimization

2025-08-30

0

PRE

AI

DOI：10.1016/j.eswa.2025.129458

Original

Original request for help

Save

Abstract

En 中文

Multiple equivalent constrained Pareto sets (CPSs) with the same constrained Pareto front (CPF) are presented in constrained multimodal multi-objective optimization problems (CMMOPs), and niches are widely used to search for equivalent CPSs in CMMOPs. However, it is crucial to select suitable niche techniques automatically for solving CMMOPs. To tackle this issue, this paper proposes a reinforcement learning-assisted automatic niche selection method, named RLANS, to automatically select a niche for solving CMMOPs. In the proposed RLANS, a reinforcement learning model is constructed and trained to obtain appropriate niche strategies in the evolutionary procedure. To begin with, diverse niche techniques are considered as the actions of the reinforcement learning model. Next, local convergence quality, feasibility, and diversity are designed and regarded as the states of the model. Then, a reward function with dynamic weight parameters is designed to evaluate and select suitable niches. Finally, RLANS trains the model using a backpropagation network and outputs the niche technique with the highest probability for selecting a suitable niche. In this way, a niche is automatically selected from the set of diverse niche technologies for locating different CPSs. The proposed RLANS and seven state-of-the-art algorithms are implemented in two standard CMMOP test suites. Experimental results validated that the proposed RLANS is able to automatically select niche techniques and exploit the advantages of different niche techniques.

Keywords:

constrained multimodal multi-objective optimization

Pareto sets

niche techniques

reinforcement learning

automatic selection

Reinforcement learning assisted automatic niche selection for constrained multimodal multi-objective optimization

Abstract

Journal

Organization