Abstract: Due to late station construction, short data series, limited measured flow data, poor quality accuracy, sudden flood surges and falls, and short convergence time, existing hydrological forecasting models for small and medium-sized rivers are difficult to achieve good forecasting and warning effects. In order to do a good job in the forecasting and early warning of small and medium-sized rivers, this article elaborates on the construction ideas and methods of multiple regression methods, as well as the application effects in the forecasting and early warning of small and medium-sized rivers in Yongzhou City, Hunan Province.It also puts forward the precautions and suggestions that need to be paid attention to in the future application process.