Simulating the Effect of Adaptivity on Randomization

Abstract

This paper compares the development of multi-chaotic system during the optimization process on three classical benchmark functions – Rosenbrock, Rastrigin and Ackley. The multi-chaotic system involves five different randomizations based on discrete chaotic maps (Burgers, Delayed Logistic, Dissipative, Lozi and Tinkerbell) and the probability of their selection is adjusted according to the development of the optimization task. Two variants of Differential Evolution (DE) are used in order to simulate the effect of adaptivity on the randomization probability adjustment process. First non-adaptive variant is DE with rand/1 mutation strategy and the second adaptive variant is novel Success-History based Adaptive DE (SHADE).