|
|
|
| Dam deformation prediction based on optimization of LSTM by using cuckoo search algorithm |
| KANG Junfeng, HU Zuochen, CHEN Youliang* |
| School of Architectural and Surveying & Mapping Engineering, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China |
|
|
|
| Guide |
|
|
Abstract Reasonable analysis and accurate prediction of dam deformation is an important means to ensure the safe operation of the dam′s safety management. The dam deformation prediction data has the characteristics of tendency, periodicity, randomness and nonlinearity. Most of the existing machine learning models are based on the nonlinear characteristics of dam deformation prediction data, while ignoring the linear characteristics of trend and periodicity of prediction data. A dam deformation prediction model was proposed by using the optimized cuckoo search algorithm(CS), the long-term and short-term memory artificial neural network(LSTM). Based on the real-time monitoring data of IoT sensors, an STL-CS-LSTM combination model was proposed. The model decomposed the dam deformation time series data into trend component, periodic component, and residual component by using the seasonal trend decomposition procedure based on the loess(STL)method of locally weighted regression. Then the optimized LSTM model was used to predict the trend component and the remainder component respectively. The simple period estimation method was used for prediction calculation. Finally, the final deformation prediction result was obtained by adding the prediction results of the three components. Lishan reservoir in Zhejiang Province was selected to carry out a deformation prediction experiment using the data of horizontal displacement and settlement automatically obtained by IoT. The results show that the STL-CS-LSTM model has the best prediction performance both in horizontal displacement and settlement deformation. The horizontal displacement prediction accuracy of other models from high to low are LSTM model, support vector regression model SVR and artificial neural network model ANN. The settlement prediction accuracy of other models is ANN model, LSTM model and SVR model.
|
|
Received: 24 January 2021
|
| Fund: |
|
|
|
|
|
|
