Enhanced Differential Evolution with Self-organizing Map for Numerical Optimization

Abstract

In Differential evolution (DE), the valuable information from the data generated during the evolutionary process has not yet fully exploited to guide the search. As a clustering algorithm based on neural network structure, Self-organizing map (SOM) method can effectively preserve the topological structure of the data in the high dimensional input space. By taking the advantage of SOM, this paper presents a SOM-based DE variant (DE-SOM) to utilize the neighborhood information extracted by the SOM method. In DE-SOM, the neighborhood relationships among the individuals are firstly extracted by the SOM method. Then, with the obtained neighborhood relationships, a self-adaptive neighborhood mechanism (SNM) is introduced to dynamically adjust the neighborhood size for selecting parents involved in the mutation process. The performance of DE-SOM has been evaluated on the benchmark functions from CEC2013, and the results show its effectiveness when compared with the original DE algorithms.