Abstract: Accurate prediction of the dam deformation during initial impoundment and a reasonable water impounding scheme are of great significance to the safe transition of a new extra-high arch dam into normal operation. To overcome the difficulties in the prediction, this paper presents a hybrid model method for the predictions of a very high arch dam, and validates it against the monitoring data of initial impounding deformation of Baihetan dam. In this method, we consider a three-stage scheme for the initial impoundment of the dam and the size of monitoring data from its crown beam section required by this scheme. It adopts two models － a multiple regression model for the first stage, and an optimized LSTM model and this regression model for five measurement points of PLdb18-2 to PLdb18-6 and the dam crown point PLdb18-1 respectively in the middle and last stages. Predictions using this hybrid model and a single model are verified against the measurements, and it shows our hybrid model method is robust and more accurate with an error lower than 4%.

Key words: hydraulic engineering, extra-high arch dam, initial impounding period, deformation, multiple regression, LSTM, hybrid model