Abstract: Medium and long-term runoff forecasting scheme is the basis for the medium and long-term dispatch plans of hydropower stations. Taking Yichang Station in the upper reaches of the Changjiang River as the main research object, based on the rainfall and runoff data of Yichang Station and related intervals from 1981 to 2019, 130 tele-correlated climate factors were introduced. The input factors were screened based on partial mutual information, and the LSTM neural network model was used to model and compare the monthly runoff forecasting schemes of Yichang Station. Studies showed that there was a strong correlation among the runoff in the upper reaches of the Changjiang River and the subtropical high pressure in the western Pacific Ocean and El Ni?o index half a year ago. In addition to tele-correlated climate factors, the runoff at Yichang Station was closely related to the runoff at Yichang Station in the same period last year, and the rainfall in the Minjiang and Tuojiang River, Wujiang River, Yalong River in the previous month. If the rainfall forecast of the Jialing River and the Yibin-Chongqing section in the current month can be applied, the accuracy of long-term runoff forecasting at Yichang Station will be greatly improved.