Deep Neural Networks for Predicting the Settlement of Earth Dams Based on the InSAR Outputs

Abstract

Earth dams play a crucial role in water resource management, necessitating effective maintenance to ensure prolonged functionality and safety. Monitoring these dams traditionally involves methods such as surveying and instrumentation. However, challenges arise from equipment malfunctions and absences, especially in dams affected by environmental changes. In response to these challenges, the cost-effective and adaptable nature of Interferometric Synthetic Aperture Radar (InSAR) has made it a preferred choice for monitoring. Despite their advantages, traditional numerical models like finite element methods have limitations in predicting deformation comprehensively, particularly due to intricate, non-linear correlations involving material type and environmental conditions. To overcome these limitations, this research employs deep learning techniques, specifically Long Short-Term Memory (LSTM) network, to capture intricate relationships and accurately predict dam behavior. Time series data from InSAR, representing settlement, are decomposed into trend and seasonal components using Artificial Neural Networks (ANN) for trend prediction. Furthermore, an LSTM network is utilized to handle the complexity of the seasonal component and its correlation with environmental factors. This network incorporates settlement, precipitation, temperature, and reservoir water level time series as inputs, thereby enhancing prediction accuracy. The research outcome presents a robust solution that holds the promise of increased accuracy and efficiency in predicting, monitoring, and serving as an early warning system for earth dam deformations over time. Such advancements are crucial for ensuring the safety and integrity of critical infrastructure in the face of evolving environmental conditions.