Article Preview
Top1. Introduction
The population-based algorithms are useful tools to deal with optimization problems such as in communication systems design (Salehinejad & Talebi, 2012; Koroupi & Salehinejad, 2012), vehicle navigation (Salehinejad & Talebi, 2010), and smart grids (Pouladi & Salehinejad, 2014). The differential evolution (DE) algorithm presents a higher performance over many other EAs because of its simplicity, effectiveness, and lower number of control parameters (Storn & Price, 1995). Over the past decade, because of its high performance, DE has been widely used to solve global optimization problems in various engineering and science fields. Similarly to other EAs, DE is a population-based stochastic algorithm which suffers from slow convergence speed. Many researchers investigated how to accelerate DE. Researches have undertaken to improve the performance of DE by control parameter fine-tuning and trial vector generation strategies for specific problems (Price & Storn, 2005), as well as obtaining those values and strategies self-adaptively (Qin & Huang, 2009). The idea of opposition-based learning can be divided into Type-I and Type-II categories (Salehinejad & Rahnamayan, 2014). The Type-I opposition learning has been employed to accelerate DE convergence speed and obtain more accurate solutions, known as opposition-based differential evolution (ODE) (Rahnamayan & Tizhoosh, 2008). By using opposition-based computation (OBC) concepts (Tizhoosh, 2005), the ODE benefits from a stronger exploration capability resulting in faster convergence. In ODE, OBC served to generate the opposite of current candidate solutions. OBC defines an opposite-point based on a predefined relationship between the extreme points (i.e., max and min) of a current population and the trial point. Thus, applying OBC to a candidate-solution generated by DE means that the opposite solution is calculated by using coordinates of the candidate solution, the maximum and the minimum of the population along each dimension as boundary points. In this paper, a new OBC scheme called centroid opposition-based computation (COBC) is proposed to meet the above features.
This is the first time where OBC is based on the entire population, by using centroids, rather than the min and max points. The centroid is the point where the centre of mass lies in a uniform body. For this matter, it is assumed that the entire population of DE is a discrete body; so the unit mass is distributed for each candidate-solution in the search space. The use of a centroid does not strengthen the learning process of an algorithm.
A new variant of ODE that uses the COBC scheme is proposed and named centroid-based opposition different evolution (CODE) (Rahnamayan & Jesuthasan, 2014). The comprehensive experimental results in this paper confirm that the CODE algorithm is faster than both ODE and DE and it also provides higher accuracy than its counterparts.