Empirical investigations on evolution strategies to self-adapt the mutation and crossover parameters of differential evolution algorithm. (12th November 2021)

• Record Type: Journal Article
• Authors: Dhanalakshmy, Dhanya M.; Jeyakumar, G.; Velayutham, C. Shunmuga
• Is Part Of: International journal of intelligent systems technologies and applications. Volume 20:Number 2(2021)
• Page Start: 103
• Page End: 125
• Publication Date: 2021-11-12
• Subjects: differential evolution -- self-adaptation -- scale factor -- crossover rate -- evolving parameters; Artificial intelligence -- Periodicals; Intelligent control systems -- Periodicals; 006.3
• Languages: English
• ISSNs: 1740-8865

Abstract

Differential evolution (DE) is a simple and robust evolutionary algorithm. Incorporation of parameter control mechanisms to DE is one thriving research area to improve its performance. Evolving the parameters along with the candidate solutions is a popular parameter control strategy. This paper proposes to investigate different parameter evolution strategies for two of its control parameters - mutation step size (F) and crossover probability (CR), using a well-defined set of four benchmarking problems with diverse characteristics and two performance metrics. This study discusses the influence of F and CR on the performance of DE, implements 25 difference instances of self-adaptive strategy for F and CR, analyses the performance to select the best strategy and validates the superiority of the proposed strategy on solving the RFID reader placement problem. The study found that the proposed strategy could solve the benchmarking functions and the chosen real-world problem faster than classical DE algorithm.