IMODEII: an Improved IMODE algorithm based on the Reinforcement Learning

The success of differential evolution algorithm depends on its offspring breeding strategy and the associated control parameters. Improved Multi-Operator Differential Evolution (IMODE) proved its efficiency and ranked first in the CEC2020 competition. In this paper, an improved IMODE, called IMODEII, is introduced. In IMODEII, Reinforcement Learning (RL), a computational methodology that simulates interaction-based learning, is used as an adaptive operator selection approach. RL is used to select the best-performing action among three of them in the optimization process to evolve a set of solution based on the population state and reward value. Different from IMODE, only two mutation strategies have been used in IMODEII. We tested the performance of the proposed IMODEII by considering 12 benchmark functions with 10 and 20 variables taken from CEC2022 competition on single objective bound constrained numerical optimisation. A comparison between the proposed IMODEII and the state-of-the-art algorithms is conducted, with the results demonstrating the efficiency of the proposed IMODEII.

I. Introduction

In many practical decision-making procedures, optimisation is a critical component of the process. Optimization has caught the attention of many scholars and practitioners for many decades because of its potential to solve planning, scientific, and technical design challenges that emerge in industry, the public sector, and the private sector. Optimisation problems identify the optimal option from a set of candidate solutions that maximises or minimises the intended results [1], [2]. These optimisation problems may be categorised in a variety of ways depending on the number and type of the involved variables, the type and number of objective functions to optimise, the presence of constraints, and a variety of other characteristics [3]. The paper's primary objective is to tackle bound-constrained optimisation problems that have a variety of mathematical features that traditional optimisation techniques cannot solve while Swarm Intelligence (SI) and Evolutionary Algorithms (EAs) techniques can.