Offline determinations of parameter values in genetic algorithm

There are two kinds of methods to determine parameters in GAs: online and offline. This paper studied the offline determinations of parameters from the decision space but not fitness landscape. In order to make full use of operators' ability to explore/exploit the subspace, the population size and terminal generation number should satisfy two conditions: (1) for each individual in the search space, the probability to be visited is greater than 0; (2) the total number of solutions that the algorithm visits should be no more than the search space size. Based on these two conditions, the upper bound of terminal generation number and the lower bound of mutation probability were given. And from the viewpoints of the subspace that crossover and mutation can cover, the value determinations for these low bound of population size and the low bound of termination generation number were proposed. The results proposed in this paper provide the theoretic basis for the application of GAs.