Neurodynamic differential evolution algorithm and solving CEC2015 competition problems

Recently, the success history based parameter adaptation for differential evolution algorithm with linear population size reduction has been claimed to be a great algorithm for solving optimization problems. Neuro-dynamic is another recent approach that has shown remarkable convergence for certain problems, even for high dimensional cases. In this paper, we proposed a new algorithm by embedding the concept of neuro-dynamic into a modified success history based parameter adaptation for differential evolution with linear population size reduction. We have also proposed an adaptive mechanism for the appropriate use of the success history based parameter adaptation for differential evolution with linear population size reduction and neuro-dynamic during the search process. The new algorithm has been tested on the CEC'2015 single objective real-parameter competition problems. The experimental results show that the proposed algorithm is capable of producing good solutions that are clearly better than those obtained from the success history based parameter adaptation for differential evolution with linear population size reduction and a few of the other state-of-the-art algorithms considered in this paper.

I. Introduction

Optimization plays an important role in many practical decision making processes. For several decades, optimization has attracted the attention of many researchers and practitioners for solving planning, scientific and engineering design problems that arise in industry, public and private sectors. Optimization problems can be classified as unconstrained or constrained optimization problems. The optimization problems may contain different types of variables, such as: integer, real, discrete, continuous or mixed [1]. In constrained problems, they may have equality or inequality constraints, or both. The objective and constraint functions can be either linear or nonlinear, convex or non-convex, continuous or discontinuous, and uni-modal or multimedia. The feasible region of such problems could be tiny compared to the search space, and it could be either one single bounded region or a collection of multiple disjointed regions. The optimal solution may or may not exist on the boundary of the feasible region of a constrained problem [2].