In this paper, we use the genetic algorithm as a tool to optimize the parameter of the model based on grey number layers and grey number interval prediction. Firstly, the mean variable weight generating thoughts is used instead of the non-bias generator to obtain the real number from the interval grey number. Then the GM(1,1) model is built according to the real number sequence. The optimal-operator is designed by minimizing the average relative error of the grey interval perdiction. The example simulation is given to illustrate that the average relative error and prediction error of the proposed model are significantly lower than that of other models, which shows the effectiveness and feasibility of the proposed model.