As renewable energy is being integrated into power system, it brings big challenges to power system, especially for online operation. Online key power flow interface automatic discovery and total transfer capability (TTC) calculation based on continuation power flow are too complex and rather time-consuming to be applied online. From the perspective of data-driven, firstly feature sets based on power flow to illustrate the power system status are established, then clustering methods are used to divide the features into several sets which fit in distributed setting. Then a feature selection method is used to select the most valuable features. Finally, artificial neural networks are utilized to map the selected features to TTC. Numerical tests show that the proposed method can significantly improve the forecasting performance of TTC and save time. This work is supported by the Science and Technology Program of China Southern Power Grid:the Key Technology and Demonstration Application for Security Feature Selection and Knowledge Discovery in Complex Large-scale Power System based on Big Data (No. GDKJ00000058).