A simple two-phase differential evolution for improved global numerical optimization

In the evolutionary computing community, differential evolution (DE) is well appreciated as a simple yet versatile population-based, non-convex optimizer designed for continuous optimization problems. A simple two-phase DE algorithm is presented in this article, which aims to identify promising basins of attraction on a non-convex functional landscape in the first phase, and starting from those previously identified search regions, a success history-based switch parameter DE is employed to further fine tune the search process leading to the optima of the landscape. Our proposed framework has been validated on the well-known IEEE Congress on Evolutionary Computation (CEC) benchmark suites (CEC 2013, 2014 and 2017). Results of the proposed method are compared with corresponding CEC winners (SHADE for CEC 2013, L-SHADE for CEC 2014 and jSO for CEC 2017).