The energy saving and consumption reduction of electrified railroad is an urgent problem, and the study of power prediction of substation-level multi-load rail transit system is of great significance to promote the research of optimal energy dispatch of railroad system. In this paper, adaptive power process detection, classification and modeling methods are proposed, and on this basis, an accurate power process forecasting method of multi-load rail transit system is proposed in combination with traffic operation diagram. The practical effect of the prediction method is analyzed with measured data, and the result shows that the forecasting method proposed in this paper can achieve high accuracy prediction of the power process of rail transit system.