High Arch Dam Displacement Prediction Model Based on Long Short-Term Memory Networks with Attention Mechanism

Dam behavior prediction that can evaluate the operational states and provide important information for safety control of dams, is an essential component of dam health monitoring. Statistical models based on regression methods have been successfully established and applied in structural health monitoring of practical engineering. However, these conventional models cannot capture the time series patterns and rely on manual parameter design. To address these problems, considering that displacement prediction is a typical time series problem, this study proposes a displacement prediction model of concrete dams using long short-term memory network (LSTM) based on deep learning techniques. The attention mechanism is adopted to capture key characteristics that influence displacement significantly. Performance of the proposed model is verified on a high arch dam. Results show that the LSTM based model outperforms the stepwise regression, back propagation neural networks, and multiple linear regression models for dam health monitoring, indicating that the proposed method is powerful and promising for arch dam behavior prediction.

Keywords: dam behavior prediction, structural health monitoring, machine learning, long short-term memory network.