Dynamic Perturbation for Population Diversity Management in Differential Evolution

Le Van Cuong, Bao Nguyen, Nguyen Khanh Phuong, Huynh Thi Thanh Binh

Abstract: Differential Evolution (DE) has achieved impressive results in solving continuous optimization problems. Like other evolutionary algorithms, the performance of DE is closely related to the population diversity. However, DE is even more sensitive to this factor than other algorithms since its mechanism of generating offspring depends wholly on the differences between individuals. This paper presents a simple perturbation technique to maintain the population diversity in which the noise intensity is adjusted dynamically during the search. A new mutation strategy called DE/target/1 and a modification of the well-known L-SHADE adaptation method are also introduced to manipulate the convergence behaviour of DE. By incorporating all the proposed techniques, we develop a new variant of DE called S-LSHADE-DP. Empirical results conducted on the benchmark suite of CEC ’22 competition show that although the ideas behind the proposals are simple, S-LSHADE-DP is highly competitive and superior in many test problems compared to state-of-the-art DE-based algorithms. The sensitivity analysis also suggests that all proposed techniques have significant contributions to the overall performance of S-LSHADE-DP.