Abstract: As a full domain searching algorithm in the area of intricate multi-apex function solution, machine learning, production scheduling, large scale combination optimizing, adapting controlling and the best scheme seeking of the skill control, the GA has an increasingly remarkable application prospect. However, the basic operation of GA is based on the randomicity of possibility. Without the measure of the evolving state, the calculation of the affinity is only dependent on the objective function with the neglection of its relations with the change of the diversity. As a result, the production of degenerated individual and repeated individual is unavoidable, which leads to the genetic excursion and premature convergence. The genetic operation put forward in this paper, with the measurement of GA's evolving state and with the elimination of the influence of the randomicity, greatly improves the searching efficiency. The experiment shows the validity of this method.