Abstract
This paper proposes a Differential Evolution based algorithm for numerical optimization in the presence of noise. The proposed algorithm, namely Noise Analysis Differential Evolution (NADE), employs a randomized scale factor in order to overcome the structural difficulties of a Differential Evolution in a noisy environment as well as a noise analysis component which determines the amount of samples required for characterizing the stochastic process and thus efficiently performing pairwise comparisons between parent and offspring solutions.
The NADE has been compared, for a benchmark set composed of various fitness landscapes tinder several levels of noise bandwidth, with a classical evolutionary algorithm for noisy optimization and two recently proposed metaheuristics. Numerical results show that the proposed NADE has a very good performance in detecting high quality solutions despite the presence of noise. The NADE seems, in most cases, very fast and reliable in detecting promising search directions and continuing evolution towards the optimum.