Long-term dam behavior prediction with deep learning on graphs

• Published in: Journal of computational design and engineering Vol. 9; no. 4; pp. 1230 - 1245
• Main Authors: He, Peng, Pan, Jianbo, Li, Yueyang
• Format: Journal Article
• Language: English
• Published: Oxford Oxford University Press 01.08.2022; 한국CDE학회
• Subjects: Arch dams; Artificial neural networks; Case studies; Dams; Deep learning; Electric power; Engineering; Graphs; Hydroelectric power; Modules; Prediction models; Set theory; Spatiotemporal data; Time series; Water resources; 기계공학; graph convolutional network; health monitoring; spatial-temporal feature; long short-term memory networks; dam displacement prediction
• ISSN: 2288-5048; 2288-4300; 2288-5048
• DOI: 10.1093/jcde/qwac051

Abstract Dam displacement prediction is one of the most crucial considerations for ensuring the dam’s long-term safe operation. Most existing models focus on predicting individual displacement and ignore the spatial and temporal correlation of data. To address these issues, a novel prediction model based on attention mechanism and graph convolutional network is proposed. To extract the spatial and temporal correlation of the original data, the position embedding and aggregation modules are employed in the prediction model. Through the aggregation module, a spatial-temporal graph is constructed. The spatial-temporal chart connects spatial diagrams of different time steps together. To capture the spatial-temporal features in the constructed graph, a recurrent graph convolutional module is employed. Through the recurrent graph convolutional module, the spatial-temporal features are used to predict the dam displacement. For verification, an arch dam is taken as an example. Comparing eight baseline models, the proposed model is more effective than other prediction models. Therefore, the proposed model can be adapted for engineering applications. Graphical Abstract Graphical Abstract