A Heuristic Repelling Strategy for Accelerating the Global Convergence of Intelligent Optimization Algorithms - A Case Study on Differential Evolution Algorithm

This work proposed a heuristic-based strategy to improve the global convergence of Intelligent Optimization Al-gorithms (IOAs) by scrutinizing their evolutionary mechanisms. Specifically, according to Gaussian and Cauchy distribution functions, six potential individuals were generated primarily based on the current best, worst, and farthest one from the best in the population. Then, they replaced the six least effective individuals, and the rest remained unchanged to form a new population to continue evolving. This process was repeated until termination. A case study on the Differential Evolution (DE) algorithm was detailed. Simulations on 21 benchmark functions, including 13 high-dimensional ones, demonstrated improved accuracy and reduced computational effort compared to some existing methods. The parameter settings also became more lenient, and the strategy showed potential for applicability to other swarm intelligence-like algorithms.

I. Introduction

The problem of global optimization (GO) remains a chal-lenging subject. Applications can be found in industry, agricul-ture, and management. Currently, GO algorithms playa critical role in deep learning (DL) techniques with the advancement of artificial intelligence, as has been shown by recent studies [1], [2].