Varying Variants for AncDE with MDV between Target and Trial Vector Measurement

Abstract

This paper compares standard Differential Evolution algorithm with AncDE, which adds a separate cache of recent ancestors that serve as an additional source of highquality genetic information. We compare the solutions produced by both DE and AncDE algorithms using benchmarks of 15 different numeric optimisation problems. Two distinct explorations are presented. The first test is distinct algorithmic variants of AncDE. The second part of this paper defines an MDV attribute and results are presented indicating some interesting differences in MDV between the DE and AncDE algorithms. Our findings indicate that ancestors can help to overcome some of the local variations in solutions quality and improve solution quality by improving population diversity.