LSTM-Based Deformation Prediction Model of the Embankment Dam of the Danjiangkou Hydropower Station

The Danjiangkou hydropower station is a water source project for the middle line of South-to-North Water Transfer Project in China. dam composed riverbed concrete and earth rock on both banks, with total length 3442 m. Once wrecked, it will yield disastrous consequences. Therefore, very important to evaluate safety behavior time. Based long-term short-term memory (LSTM) network, deformation prediction models embankment are constructed. contain two LSTM layers, adopt rectified linear unit function as activation determine super parameters Bayesian optimization algorithm. According settlement monitoring data LD12ZT01 measuring point (dam crest 0 + 648) left bank from July 2013 March 2022, bidirectional In total, 80% taken training set 20% test data. mean absolute error, root square error 0.42978, 0.56456 0.31873 partial least squares regression (PLSR), 0.35264, 0.47561 0.22621 0.34418, 0.45400 0.20612 LSTM, respectively. results show that model can obtain better value than PLSR. Then, used predict LD16YT01 658) right bank, 0.5425, 0.66971 0.4520, This shows effectively station.