Loading...

Predicting the Behavior of a Double-Arched Concrete Dam Using the Machine Learning Technique

Moradi Sarkhanlou, Milad | 2023

14 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56921 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Ghaemian, Mohsen; Toufigh, Vahab
  7. Abstract:
  8. In recent years, machine learning techniques have been available to predict and interpret the structural behavior of dams. The main goal of this research is to predict structural behavior through modeling with data read by accurate instruments of dams. This requires the selection of several machine learning methods such as Random Forests (RF), Boosted Regression Trees (BRT), Artificial Neural Networks (ANN), Support Vector Regression (SVR), and Decision Tree Regression (DTR) algorithms. In this research, considering the 24 target variables defined from the detailed instrumentations studied, the behavior of the dam is modeled using selected machine learning models. These precision instruments include Pendulum-Radial displacement, Pendulum-Tangential displacement, Joint meter, and Extensometer. Each of these precision instrumentations is independently modeled and evaluated for 6 target variables. To ensure that the constructed models align with the data, it is imperative to validate and verify them. Consequently, the evaluation of the models is carried out in two phases. In the first phase, the accuracy of the models in the training dataset is assessed using three error measurement indices (MAE, MSE, and R2). The second phase involves the implementation of validation techniques and model validation on both training and prediction datasets. These validation techniques encompass historical data validation, predictive validation, and the examination of the time evolution of residuals. In this way, the evaluation and validation process led to the selection of the BRT model as the best model, with an accuracy ranging from 0.95 to 0.99 in the R2 function and the best adaptation and accuracy in both training and predicting the behavior of the studied arch dam in all four displacements. Therefore, the selected model for predicting the structural behavior of the arch dam for the next two years requires data related to the input time series variables (such as reservoir level changes, air temperature changes, etc.) defined by the model. For this reason, the long short-term memory (LSTM) model, which is a robust algorithm for predicting time series variables, has been used to predict these variables. The results show that the LSTM model has provided acceptable predictions of possible changes in the input variables in the next two years of the dam. Also, the BRT model, selected as the most accurate model in the evaluation process, was able to provide appropriate predictions of the structural displacements of the arch dam for periods not yet experienced by the dam using these possible input variables. The safety evaluation of the displacement prediction results shows that the dam exhibits uniform and regular behavior in the central blocks. However, towards the left and right side blocks of the dam, the displacements are increasing, indicating the presence of injuries and damages such as corrosion cracks in the side blocks of the dam
  9. Keywords:
  10. Instrumentation ; Concrete Arc Dam ; Machine Learning ; Validation ; Double-Arch Concrete Dam ; Behavior Prediction ; Safety Control

 Digital Object List

 Bookmark

No TOC