Aiming at the optimization problem of the weight and the threshold value in the prediction of satellite clock bias by BP (Back Propagation) neural network model, a short-term prediction model of satellite clock bias based on genetic algorithm is proposed, the basic idea, the specific method, and the implementation steps of genetic algorithm optimization of BP neural network are presented. In order to verify the validity and feasibility of the optimization model, the accuracy of clock bias prediction is analyzed by using the data of BDS (BeiDou navigation satellite system) satellite clock bias, and the prediction precisions are compared with the GM(1,1) (Grey Model(1,1)) and BP neural network model. The results show that the prediction precision for the proposed method is better than those for the GM(1,1) and BP neural network models.