Abstract:This article proposes a combined Long Short Term Memory Network(LSTM) and K-Nearest Neighbor (KNN) model for predicting hydrological flow, and analyzes the daily runoff of the Jimai,Jungong and Tangnaihai hydrological sections in the source area of the Yellow River.First,the dynamic attributes of the watershed are constructed by using meteorological factors such as temperature and precipitation,Using historical meteorological and hydrological and geographic information data to build the static attribute characteristics of river basin,The LSTM model is used for feature optimization,Determine the optimal model TOPO_ CLIM_ SOIL-LSTM, it is used for actual daily runoff forecast,Then the KNN algorithm is used to correct the prediction results in real time.The results show that,Model TOPO_CLIM_SOIL-LSTM with river basin attribute characteristics can better learn the relationship between rainfall and runoff,It can effectively solve the problem of predicting runoff jump in low flow section,After revising the predicted flow with KNN model,The accuracy rate of daily runoff forecast at Jimai,Jungong and Tangnaihai stations in the next day has reached over 93%,Nash coefficient increased by 6.45%,12.5% and 18.07% respectively,Effectively improve the accuracy of daily runoff forecast.