DSM-DE: a differential evolution with dynamic speciation-based mutation for single-objective optimization
- PDF / 1,420,102 Bytes
- 14 Pages / 595.276 x 790.866 pts Page_size
- 1 Downloads / 168 Views
REGULAR RESEARCH PAPER
DSM-DE: a differential evolution with dynamic speciation-based mutation for single-objective optimization Libao Deng1
· Lili Zhang1 · Haili Sun1 · Liyan Qiao1
Received: 12 April 2018 / Accepted: 3 January 2019 © Springer-Verlag GmbH Germany, part of Springer Nature 2019
Abstract A new differential evolution algorithm with two dynamic speciation-based mutation strategies (DSM-DE) is proposed to solve single-objective optimization problems. An explorative mutation “DE/seeds-to-seeds” and an exploitative mutation “DE/seeds-to-rand” are employed simultaneously in DSM-DE in the evolutionary process. A Dynamic Speciation Technique is designed to assist the two mutations in order to utilize the potential of selective portioning of critical individuals in the population. It dynamically divides the population into numbers of species whilst taking species seeds as centers. The best individuals for each species are used as base vectors in each species in the proposed mutation strategies. “DE/seeds-to-seeds” selects individuals from species seeds and current species to constitute difference vectors whereas “DE/seeds-to-rand” selects from the whole population. Thus the two mutation strategies can accelerate the convergence process without decreasing diversity of the population. Comparison results with four classic DE variants, one state-of-art DE variant and two improved non-DE variants on CEC2014, CEC2015 benchmark, and Lennard-Jones potential problem reveal that the overall performance of DSM-DE is better than that of the other seven DE algorithms. In addition, experiments also substantiate the effectiveness and superiority of two seeds-guided mutation strategies in DSM-DE. Keywords Differential evolution · Mutation strategy · Dynamic speciation · Single-objective optimization
1 Introduction Differential evolution (DE) is a simple yet powerful algorithm firstly proposed by Storn and Price in 1997 [27]. It has attracted extensive attention of scholars to find new variants because of its excellent performance and has been used in various engineering fields [8,23]. DE is a population-based stochastic search technique employing mutation, crossover, and selection operators at each generation to evolve the population to the global optimum. The classic DE employs “DE/rand/1/bin” mutation in which three parent vectors are randomly chosen from the current population. It is of robust capacity in exploring the whole solution space and locating the global optimal region but of less efficient convergence rate when exploiting the optimal solution. Greedy This work was supported in part by National Natural Science Foundation of China (61401121).
B 1
Libao Deng [email protected] School of Information Science and Technology, Harbin Institute of Technology, Weihai, Shandong, China
mutation strategies such as “DE/current-to-best/bin” and “DE/best/bin”, which utilize the best solution information in the current population, usually display higher convergence rate but less reliable. However, the reliability of
Data Loading...
