A reinforcement learning-based metaheuristic algorithm for solving global optimization problems. (April 2023)

• Record Type: Journal Article
• Authors: Seyyedabbasi, Amir
• Is Part Of: Advances in engineering software. Volume 178(2023)
• Publication Date: 2023-04
• Subjects: Metaheuristic algorithm -- Reinforcement learning algorithm -- Sand cat swarm optimization -- Q-learning -- Machine learning; Computer-aided engineering -- Periodicals; Engineering -- Computer programs -- Periodicals; Engineering -- Software -- Periodicals; Periodicals; 620.0028553
• DOI: 10.1016/j.advengsoft.2023.103411
• Languages: English
• ISSNs: 0965-9978

Abstract

Highlights: Metaheuristic algorithms find optimal solutions to global optimization problems in random search spaces. It has been shown that reinforcement learning methods are more successful in finding new global areas than metaheuristic approaches, and have a more balanced behavior than metaheuristic methods. In this paper, we propose an algorithm that switches exploration and exploitation phases effectively using Q-table values in the RLSCSO algorithm. Several algorithms have been tested on a number of well-known benchmarks as well as the 100-Digit Challenge on Single Objective Numerical Optimization functions and have been applied to the problem of localizing mobile sensor nodes. Abstract: The purpose of this study is to utilize reinforcement learning in order to improve the performance of the Sand Cat Swarm Optimization algorithm (SCSO). In this paper, we propose a novel algorithm for the solution of global optimization problems that is called RLSCSO. In this method, metaheuristic algorithm is combined with reinforcement learning techniques to form a hybrid metaheuristic algorithm. This study aims to provide search agents with the opportunity to perform efficient exploration of the search space in order to find a global optimal solution by using efficient exploration and exploitation to find optimal solutions within a given search space. A comprehensive evaluation of the RLSCSO has been conducted on 20 benchmark functions and 100-digit challenge basic test functions. Additionally, the proposed algorithm is applied to the problem of localizing mobile sensor nodes, which is NP-hard (nondeterministic polynomial time). Several extensive analyses have been conducted in order to determine the effectiveness and efficiency of the proposed algorithm in solving global optimization problems. In terms of cost values, the RLSCSO algorithm provides the optimal solution, along with tradeoffs between exploration and exploitation.