Deformation prediction of a super-high arch dam during initial impoundment based on the HST-NN integrated model

Abstract

During the initial impoundment period of super-high arch dams, the monitoring data are characterized by dynamic variation and complex environmental influences. Traditional models often suffer from poor prediction stability under data-scarce conditions and insufficient capability to capture nonlinear patterns when data are abundant. Therefore, there is an urgent need to establish a high-adaptability prediction model that integrates physical constraints and data-driven capabilities. This paper proposes an HST-NN integrated model, which combines the physical mechanism constraints of the HST(Hydrostatic-Seasonal-Time) formula model with the nonlinear feature learning ability of a Deep Neural Network(DNN), to accommodate the evolving data conditions throughout the entire impoundment process. The model is constructed and validated in stages based on measured deformation data from the initial impoundment of a 300 m-class super-high arch dam.Results show that the proposed model maintains a relative error within 10% under data-scarce conditions and further reduces to below 5% as data increases. Compared to traditional single models, the HST-NN model significantly improves prediction accuracy and stability. It demonstrates strong adaptability and generalization, providing an effective tool for deformation prediction and safety management during the initial impoundment of super-high arch dams.

Keywords

super-high arch dam / initial impoundment / deformation prediction / HST-NN model / deep learning