Loading...

Feasibility Study for the Prediction of Earth Dam Behavior Using Combined Methodology of Deep Learning and Remote Sensing

Babaei Ghareh Tappeh, Amir Hossein | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57238 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Sadeghi, Hamed
  7. Abstract:
  8. Earth dams are among the most important and widely used civil structures, playing a significant role in water supply, energy production, and flood control. Monitoring dam behavior provides professionals with insights into its performance over time. Dam displacement is a crucial factor in deformation and potential failure, making it essential to use various measurement tools to control deformation and numerous simulation models to analyze data and monitor the dam continuously. However, it is not always possible to record data regularly and without disturbance, and some dams may lack these measurement tools altogether. In such cases, new technologies like remote sensing can significantly aid in monitoring the structural health of dams. Additionally, the use of intelligent models to analyze measured data enhances the accuracy of dam monitoring. The advancement of artificial intelligence methods in data analysis has reduced computational and time costs, enabling faster and more precise monitoring. This research aims to utilize Sentinel-1 satellite images to measure the displacement of the Karkheh Dam and predict it using the LSTM deep learning model. To investigate the impact of environmental factors, including temperature, precipitation, and reservoir water level, on dam behavior, the most critical displacement point at the dam crest, shows a settlement of 8.2 cm with less than a 2 cm difference from the dam's instruments, was analyzed using the STL algorithm. The correlation coefficients of its trend and seasonal components with the mentioned environmental parameters were calculated. Temperature (T) and precipitation (P) showed a non-linear relationship with seasonal displacement, while the reservoir water level (W) had a significant linear relationship with total displacement. Therefore, these parameters were incorporated into the seasonal displacement (S) forecasting model to improve results. Eight combinations, including S, SP, ST, SW, SPT, SPW, STW, and SPTW, were created for the LSTM model input. Incorporating environmental data into the predictive model increased accuracy. In a way that, using the time series of temperature and precipitation along with seasonal displacement (SPT) reduced the MSE criterion from 0.52 to 0.16. Temperature and precipitation had the most significant effect on improving forecast accuracy, while the reservoir water level decreased the forecast accuracy
  9. Keywords:
  10. Remote Sensing ; Deep Learning ; Artificial Intelligence ; Earth Dam ; Monitoring ; Health Monitoring

 Digital Object List

 Bookmark

...see more