SWLSTNet-GPR: a novel point and probabilistic prediction model for concrete dam displacement based on deep learning algorithms

Establishing an accurate and efficient prediction model for concrete dam deformation is crucial for ensuring dam safety. However, most existing studies primarily focus on enhancing point prediction accuracy, often neglecting important factors, such as training efficiency and probabilistic forecastin...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Journal of civil structural health monitoring Ročník 15; číslo 8; s. 3361 - 3381
Hlavní autori: Zhang, Ye, Zhang, Wenwei, Li, Yanlong, Wen, Lifeng, Sun, Xinjian
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: Berlin/Heidelberg Springer Berlin Heidelberg 26.07.2025
Predmet:
Civil Engineering
Control
Dynamical Systems
Engineering
Measurement Science and Instrumentation
Original Paper
Vibration
Deep learning
GPR
Deformation
Hydraulic Engineering
SWLSTNet
Probabilistic prediction
ISSN:2190-5452, 2190-5479
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:Establishing an accurate and efficient prediction model for concrete dam deformation is crucial for ensuring dam safety. However, most existing studies primarily focus on enhancing point prediction accuracy, often neglecting important factors, such as training efficiency and probabilistic forecasting. To address this gap, a hybrid concrete dam deformation prediction model based on SWLSTNet-GPR is proposed in this study. This model integrates the SWLSTM module within the Long- and Short-term Time-series Network (LSTNet) framework, significantly reducing training time while effectively balancing training efficiency and prediction accuracy. Additionally, Gaussian Process Regression (GPR) is employed to generate probabilistic forecasts, allowing for the quantification of uncertainty in the prediction process. The model is validated using monitoring data from a concrete gravity dam located on the Tibetan Plateau. The experimental results indicate that the SWLSTNet-GPR model consistently achieves the shortest training time and the highest prediction accuracy, with the coefficient of determination ( R 2 ) always exceeding 0.97. Compared with the original LSTNet, the training time is reduced by at least 30% while maintaining reliable probabilistic predictions. Overall, this research provides valuable insights for risk assessment and decision-making in dam safety management and shows promise for broader application to other structural behavior prediction tasks.
ISSN:2190-5452
2190-5479
DOI:10.1007/s13349-025-00994-y