Toward explicit control between exploration and exploitation in evolutionary algorithms: A case study of differential evolution

Exploration and exploitation are two cornerstones of evolutionary algorithms. An appropriate balance between exploration and exploitation can drive a search process toward global optima with a fast convergence rate. However, this balance is not comprehensively understood, and the issue of how to effectively control it is very challenging. In this paper, a new search framework based on an explicit control strategy that balances the amounts of exploration and exploitation in a search process is proposed. First, an explicit control strategy consisting of three types of transference is proposed to balance exploration and exploitation. Then, exploration and exploitation operators are formally defined by adaptive Gaussian local search with reinitialization and multioffspring-based differential evolution, respectively. Finally, a new triple-transference-based differential evolution method is proposed. The experimental results on 29 benchmark optimization functions show its outstanding performance, especially on complex problems. The balance between exploration and exploitation in the proposed algorithm is discussed in detail. The success of this new framework provides more insights into the principles of balancing exploration and exploitation. It also leads us to believe that exploration and exploitation in evolutionary algorithms can eventually be explicitly controlled.