Self-adaptive Differential Evolution Algorithm Based On Computational Verb Rules

Handing multimodal functions in continuous space is a very important and challenging task in evolutionary computation community. To search global optimal of upon functions, we use Differential Evolution (DE) algorithm, which is a simple and powerful population-based stochastic search technique. Butfor high dimension complex multimodal functions stochastic search shows instability properties in optimization process. Therefore, an efficient guidance to interesting area during evolution is a worth studying subject.Computational Verb Theory is a fuzzy dynamical system with property evolution, which follows several fuzzy logic inference rules. Motivated by the dynamics and the proximity characteristics of Computational Verb Rules, we propose two approaches for self-adaptive DE, which lead the evolution process to better solutions in reasonable ways.The first approach is combining computational verb rules with niching mutation strategy. The niching method is presented to avoid the premature phenomenon in DE, aiming to maintain the diversity of evolution in each generation, which preventing the convergence to local optima. Since the conventional DE is sensitive to the setting of parameters, the second approach is structuring a two layered DE with self-learning control parameters, and the evolution process of parameters are directed by computational verb rules which will lead to a better fitting parameters.Both two proposed modifications are tested on commonly used benchmark problems for unconstrained optimization. Experimental results with some state-of-art algorithms indicate that the proposed combination of computational verb rules is competitive and very promising. It improves the efficiency and robustness of the algorithm.