Parameter control by the entire search history: Case study of history-driven evolutionary algorithm
Shing Wa Leung, Shiu Yin Yuen, Chi Kin Chow · 2010
History-driven Evolutionary Algorithm (HdEA) is an EA that uses the entire search history to improve searching performance. By building the approximated fitness landscape and estimating the gradient using the entire history, HdEA performs a parameter-less adaptive mutation. In order to decrease the number of parameters that makes the HdEA more robust, this paper proposes a novel adaptive parameter control system. This system is as an add-on component to HdEA, which uses the whole search history in HdEA to control the parameters in an automatic manner. The performance of the proposed system is examined on 34 benchmark functions. The results shows that the parameter control system gives similar or better performance in 24 functions and has the benefit that two parameters of the HdEA are eliminated; they are set and varied automatically by the system.