Applying exponential weighting moving average control parameter adaptation technique with generalized differential evolution

Abstract

In this paper, an Exponential Weighting Moving Average (EWMA) control parameter adaptation technique is tested with Generalized Differential Evolution 3 (GDE3) using a set of multi-objective test problems and performance metrics. The results with and without EWMA control parameter adaptation are compared. EWMA has been earlier proposed with the original unconstrained single-objective Differential Evolution (DE), and EWMA adapts crossover and mutation control parameter values. From the results, it is observed that if good initial control parameter values are used, then there is not clear performance difference between the original GDE3 and GDE3 with EWMA. However, if the initial control parameter values are not good, then EWMA gives clear improvement in performance. Since GDE3 with EWMA is identical with the original DE in the case of single-objective optimization, the same control parameter adaptation technique can be used both in the case of single- and multi-objective optimization. However, different initial control parameter values should be used in different cases, and recommendations for initial values are given at the end of the paper.