PERFORMANCE ENHANCEMENT OF DIFFERENTIAL EVOLUTION BY DIRECT ALGORITHM

Differential Evolution (DE) is one of the effcient Evolutionary Algorithms (EAs) for the continuous optimization domain. Similar to other EAs, DE algorithm works on a population of candidate solutions. Population initialization in DE is an important operation because it can affect the convergence speed and also the quality of the obtained solution. If we can provide good initial population to DE, improvement of search efficiency can be expected. However, in black-box optimization, there exists no prior information about the search landscape of a given problem. Therefore, in this research we use DIRECT (Dividing RECTangles) algorithm to provide a good initial population to DE. DIRECT is a deterministic global optimization algorithm for boundconstrained problems. The algorithm, based on a space-partitioning scheme, performs both global exploration and local exploitation by one tuning parameter. In the proposed method, named DE-DIRECT, first the search by DIRECT is performed until a certain number of iterations. Next, the solutions obtained by DIRECT are used as the initial individuals of DE. The remaining search is performed by DE using DIRECT's solutions until the total budget is exhausted. In order to extract effective individuals for DE from the solution set of DIRECT, we introduce a selection method considering diversity as well as accuracy of solutions. The effectiveness of the proposed DE-DIRECT is examined and discussed by experiments on various benchmark functions.