On Adaptive Operator Probabilities in Real Coded Genetic Algorithms

In this paper a decision making scheme devised by Lobo and Goldberg for a hybrid genetic algorithm is extended to deal with the problem of adaptation of the operator probabilities of a real coded steady-state genetic algorithm applied to optimization problems. The scheme is modi ed by introducing a global reference value for measuring operator productivity as well as by the inclusion of operator ancestry when distributing rewards. The e ect of those modi cations is examined by means of numerical experiments performed on three test problems.