Noise adaptive spatiotemporal neural networks for deformation prediction of high rockfill dams

Deformation prediction is a crucial approach in structural health monitoring of high rockfill dams, significantly contributing to their construction and operational safety. However, the monitoring data of high rockfill dams exhibits significant volatility due to the influence of construction process...

Celý popis

Uložené v:

Podrobná bibliografia
Vydané v:	Expert systems with applications Ročník 294; s. 128836
Hlavní autori:	Wang, Zijian, Ma, Gang, Ai, Zhitao, Ding, Qianru, Zhou, Wei
Médium:	Journal Article
Jazyk:	English
Vydavateľské údaje:	Elsevier Ltd 15.12.2025
Predmet:	Parameter shared encoder-decoder Probabilistic prediction Rockfill dams Spatiotemporal fusion Structural health monitoring Structural health monitoring Parameter shared encoder-decoder Rockfill dams Spatiotemporal fusion Probabilistic prediction
ISSN:	0957-4174
On-line prístup:	Získať plný text
Tagy:	Pridať tag Žiadne tagy, Buďte prvý, kto otaguje tento záznam!

Popis
Shrnutí:	Deformation prediction is a crucial approach in structural health monitoring of high rockfill dams, significantly contributing to their construction and operational safety. However, the monitoring data of high rockfill dams exhibits significant volatility due to the influence of construction process and water level fluctuations, posing challenges for data-driven deformation prediction. This study proposes a novel deformation prediction model based on spatiotemporal fusion neural network for enhancing prediction accuracy and robustness to noise. Specifically, the graph convolutional network and Long Short-term memory Network are combined to converge multi-point spatial features and excavate historical temporal information. Then, probabilistic prediction method obtains parameters through the linear layer for loss function, which improves model adaptation to data noise. The parameter shared Seq2Seq structure is designed to enhance the model’s ability to learn the correlation between loads and settlement, thereby enabling accurate prediction of deformation monitoring data with drift characteristics. Eventually, by incorporating the above model structure, loss function and training strategy, deformation prediction is accomplished. Application on a high rockfill dam shows that the proposed model realizes more accurate deformation prediction, with an error (MAE) less than 1.0c m, outperforming various conventional prediction methods. The reliability of the noise adaptation module and the application value for anomaly detection are further validated, which provides a valuable reference for data-driven deformation prediction methods in rockfill dams and other geotechnical engineering projects. The code is available at https://github.com/WHU-Wzj/Dam-deformation-prediction.
ISSN:	0957-4174
DOI:	10.1016/j.eswa.2025.128836