Abstract: This paper focuses on how to analyze and model accurately the tailwater level characteristics of hydropower stations, so as to achieve high accuracy predictions. First, we reveal the aftereffect characteristics of tailwater level variations at hydropower stations by combining qualitative and quantitative analysis, and explore preliminarily the key influencing factors of the variations based on a Pearson correlation analysis. Then, we construct a polynomial fitting model and a support vector regression model, and compare and analyze their performances in prediction of tailwater level variations Case studies of the Xiluodu-Xiangjiaba cascade and Three Gorges-Gezhouba cascade show a significant aftereffect is produced by a two-hour variation in the tailwater stages of the four hydropower stations. And the support vector regression model with reservoir discharge of the present and previous periods and reservoir tailwater stage or downstream tributary discharge as inputs is a practical, reliable and accurate in prediction of the tailwater levels.

Key words: hydropower station, tailwater level characteristics, tailwater level prediction, Pearson correlation analysis, polynomial fitting, support vector regression.