Based on analysis of the deficiencies of the BP neural network used in prediction, the time series prediction based on BP neural network was discussed. According to the characteristics of BP network structure and Z transform theory, function y=x was put forward to be the transfer function in this kind of prediction. Besides, the conclusion that function y=x and y=a+bx were equivalent in the BP network was proposed. It was pointed out that the layer of network structure should be two. On basis of this, the corresponding formula was derived. With the unipolar Sfunction and function y=x as the transfer function respectively, the total power of agriculture machinery was calculated. The results showed that the performance of function y=x in BP neural network was better than Stransfer function in external push effect and influences of training sample data processing interval. It also overcame the shortcomings of Sfunction used in BP neural network. 