Control the Diversity of Population with Mutation Strategy and Fuzzy Inference System for Differential Evolution Algorit
- PDF / 2,812,410 Bytes
- 14 Pages / 595.276 x 790.866 pts Page_size
- 10 Downloads / 199 Views
Control the Diversity of Population with Mutation Strategy and Fuzzy Inference System for Differential Evolution Algorithm Jing-Zhong Wang1 • Tsung-Ying Sun1
Received: 3 January 2020 / Revised: 10 February 2020 / Accepted: 17 February 2020 Taiwan Fuzzy Systems Association 2020
Abstract This paper displays how to use fuzzy inference system (FIS) to control the individual uniform diversity for differential evolution algorithm (DE). DE solves nonlinear optimization problems, and a successful control mechanism for population diversity enhances the performance of DE. This study proposed a control mechanism that contains a novel mutation strategy and FIS because FIS is suitable for consecutive and hard classified inputs. The proposed control mechanism does not fix the target vector and controls the ratio of mutating toward the whole best individual by FIS. The FIS decides the F values for this novel mutation strategy. The experiments compared the winner of each evaluated functions among four uniform diversity goals (UDGs) with conventional strategies. From experimental results, the proposed method finds superior solutions to conventional mutation strategies at least 11 out of 15 evaluated functions in 10, 30, and 50 dimensions. Furthermore, not only the diversity curves confirm the control ability of FIS, but also different paths of convergence curves indicate the fast convergence and mitigation of evolutionary stagnation. Keywords DE Diversity control Fuzzy inference system Mutation strategy Entropy
& Tsung-Ying Sun [email protected] Jing-Zhong Wang [email protected] 1
Department of Electrical Engineering, National Dong Hwa University, Hualien, Taiwan
1 Introduction Evolutionary algorithms (EAs) solve nonlinear optimization problems. When studying EAs, how to balance exploitation and exploration during evolution is crucial. If an EA overly emphasizes exploitation, there is a high probability that it will fall into a local optimum or converge prematurely. By contrast, if an EA overly emphasizes exploration, its convergence time may become very long [1]. Numerous studies have focused on this tradeoff. Bosman and Thierens [2] proposed a framework to balance exploitation and exploration. Chen et al. [3] devised a selection strategy that included five rules to keep the genetic diversity for enhancing evolutionary programming. Segura et al. [4] define a distance metric for measuring diversity and considered the evolutionary stoppage criteria when setting their diversity goal. The differential evolution (DE) algorithm [5], one of the EAs, also suffers from the problem of balancing exploitation and exploration. Studies have focused on ensuring diversity. Yu et al. [6] used the DE/lbest/1 mutation strategy, which achieves multiple local best individuals leading instead of global best leading, to increase diversity. Another approach is to apply different mutation strategies after dividing a population into different classes. Das et al. [7] handled three individual groups—DE, Brownian, and quantum indivi
Data Loading...
