AI-based Early Warning for safe operation of large dams

A nested deep neural network-based correlation analysis for risk-forecasting was developed as the core module of an early warning system.

One of the subprojects of the DAMAST project ( is dedicated to the study of the context of hydropower, which is a significant part of the energy mix in many countries and crucial for the realization of sustainable and renewable energy systems. However, hydropower infrastructure can be vulnerable to natural hazards such as earthquakes, landslides, and heavy precipitation, which can induce seismic activity and threaten both the safety of nearby populations and the integrity of the dam itself. With climate change exacerbating the risk of extreme weather events, it is critical to develop robust early warning systems and monitoring technologies to support safe dam operation and ensure the resilience of energy systems. The focus of this sub-project was to apply sophisticated measurement technologies that enable accurate and timely monitoring, early warning, and risk assessment, thereby increasing the lifetime of dams and protecting both the population and the environment.

The RESIS department at the Institute of Thermal Energy Technology and Safety (ITES) had a specific focus on developing an early warning methodology that would enable dam and hydro plant operators to implement risk-reducing measures in a timely manner, particularly in the scenario of an extreme precipitation event. The primary element at risk in such a situation is the dam, and the early warning system's main module anticipates any potential critical deformation. By incorporating fragility models and additional environmental data, the system is also able to conduct risk assessments.

The deformation of the dam was measured using Ground-Based Synthetic Aperture Radar (GB-SAR), by the Institute of Photogrammetry and Remote Sensing (IPF), and the results of the analysis were used to develop a new nested deep neural network-based correlation analysis (DNN-A) for the prediction module. This analysis produced several findings supporting the assumption that meteorological parameters and the gradient of water level changes can both have an impact on induced seismicity, in addition to hydro-static pressure.

Based on these findings, it is suspected that induced seismicity, along with hydro-static pressure and meteorological parameters, may also have an impact on dam deformation. As a result, the DNN-A can be utilized as a core module for risk forecasting, and appropriate measures can be calculated via simulation and optimization. The team plans to further refine the system based on these findings and continue to improve the accuracy of the early warning methodology.

Abb.: Geometric decomposition of the dam surface into geodesic rectangles. We see those that contain measurement points regarding the dam deformation.

Associated institute at KIT: Institute of Thermal Energy Technology and Safety (ITES)
Author: Sadeeb Simon Ottenburger (May 2023)