Controlling of Local search Methods’ Parameters in Memetic Algorithms Using the Principles of Simulated Annealing

Abstract

In simulated annealing the probability of transition to a state with worse value of objective function is guided by a cooling schedule. The more iterations are spent, the more strict the acceptance probability function becomes. In the end of the optimization process, the probability of transfer to worse state approaches zero. In this paper the principles of cooling schedules are used to control the parameters of local search methods in the memetic algorithm. The memetic algorithm in this paper is a combination of genetic algorithm, Hooke-Jeeves method, Nelder-Mead simplex method and Dai-Yuan version of nonlinear conjugate gradient method. The controlled parameter of Hooke-Jeeves method is the radius r, for Nelder Mead method the size of edge of the simplex and the length of step for nonlinear conjugate gradient method.