Neighborhood guided differential evolution

Abstract

Differential evolution (DE) relies mainly on its mutation mechanism to guide its search. Generally, the parents involved in mutation are randomly selected from the current population. Although such a mutation strategy is easy to use, it is inefficient for solving complex problems. Hence, how to utilize population information to further enhance the search ability of the mutation operator has become one of the most salient and active topics in DE. To address this issue, a new DE framework with the concept of index-based neighborhood, is proposed in this study. The proposed framework is named as neighborhood guided DE (NGDE). In NGDE, a neighborhood guided selection (NGS) is introduced to guide the mutation process by extracting the promising search directions with the neighborhood information. NGS includes four main operators: neighborhood construction, neighbors grouping, two-level neighbors ranking, and parents selection. With these four operators, NGS can utilize the topology and fitness information of population simultaneously. To evaluate the effectiveness of the proposed approach, NGS is applied to several original and advanced DE algorithms. Experimental results have shown that NGDE generally outperforms most of the corresponding DE algorithms on different kinds of optimization problems.

Differential Evolution With Adaptive Guiding Mechanism Based on Heuristic Rules

“…In DE with multiobjective sorting based mutation operators (MSDE) [20], the multi-objective non-dominated sorting method is introduced into DE to calculate the probability of selection as the parents for all the vectors based on their fitness values and diversity indexes. In neighborhood guided DE (NGDE) [13], the ring topology is used to define the neighborhood of each vector, and the selection probability for each neighbor is calculated based on its fitness value.…”

“…It makes these strategies be good at local exploration but easily lead to premature convergence. Based on these considerations, various approaches have been proposed to enhance the search ability of the mutation operator for different complex problems, which roughly fall into the following categories: designing new mutation strategies [21], [22], integrating multiple mutation strategies [9], [10], [23], and selecting parent vectors for mutation [13]- [15], [24]. These works related to the mutation operator of DE will be reviewed in Section III.…”

Neighbor-Induction and Population-Dispersion in Differential Evolution Algorithm

“…Various studies have been conducted on neighborhood strategies to enhance the performance of the algorithm. The neighborhood strategies are usually divided into static neighbor strategies and dynamic neighbor strategies [25] according to whether a neighbor changes along with the search process.…”

“…Gong et al [36] combined a biogeographic-based migration operator into DE to guide each individual to learn from good neighbors based on their mobility. Cai et al [25] proposed a neighborhood-adaptive DE algorithm in which an index-based neighborhood topology pool is used to define multiple neighbor relationships for each individual, and self-adapting neighborhood relationships can be used in different evolution stages. Liang et al [37] introduced the fitness Euclidean-distance ratio (FER) technique into DE to locate peaks of functions.…”

Improved Parameter Identification Method for Envelope Current Signals Based on Windowed Interpolation FFT and DE Algorithm

“…Differential Evolution is a simple and efficient algorithm for global optimization over continuous spaces [25,26]. Let f (X) be the target function, the crossover rate is C, the scaling factor is F and the evolution generation is t. The steps of DE are as follows [27].…”