Evolving solutions in evolutionary optimization based on a dynamic system model

Due to stochastic nature, the success of evolutionary methodologies in search for solutions to optimization problems depends greatly on the harmonization between exploration and exploitation. However, the difficulties in compromising on these two properties properly make the design of an efficient evolutionary algorithm (EA) become one of the laborious tasks. In this paper, a novel EA based on a dynamic system model is proposed. The algorithm composes two evolutionary stages, namely attraction and mutation. Using information about location and fitness of the best individual within the current population, the attraction stage guides candidate solutions toward new locations closer to the best according to a dynamic-system-model-based guidance technique, which is designed to be favorable for exploitation yet retain local diversity. In an opposite manner, the mutation stage, whose role is much the same as in many other EAs, probabilistically changes the values of individuals' elements so as to increase the population diversity and promote the exploration of potential search areas. An early version of the proposed approach was applied to real-valued optimization on a number of benchmark functions. Preliminary results demonstrate that the algorithm is promisingly effective in finding near-optimal solutions to problems where a trade-off between the exploration and exploitation is a must.