-
- H13N-1240: Physics-Informed and Interpretable LSTM for Hydropower Reservoir Release Modeling Under Climate Variability
-
Board 1240‚ Hall EFG (Poster Hall)NOLA CC
Author(s):Generic 'disconnected' Message
Zhaoyu Zhang, The University of Tokyo (First Author, Presenting Author)
Akiyuki Kawasaki, The University of Tokyo
Abdul Moiz, Center for Western Weather and Water Extremes (CW3E), Scripps Institution of Oceanography, University of California San Diego
In mountainous regions, reservoirs are essential for storing water and generating electricity. Climate change has made it harder to manage these systems due to changing rainfall and temperature patterns. Traditionally, reservoir operations have followed experience-based rules, which may not work well under future climate conditions. Our research introduces a new framework that combines physical modeling and machine learning. We use physical models to convert weather data into hydrological variables—such as snow accumulation, snow depth, and soil moisture—which then train a machine learning model to predict reservoir operations. To make the model transparent, we apply a technique called Integrated Gradients to understand which inputs most influence the model’s predictions for power generation decisions. This approach improves the accuracy and interpretability of reservoir operation modeling and offers a more adaptable strategy for managing water and energy in a changing climate.
Scientific DisciplineSuggested ItinerariesNeighborhoodType
Enter Note
Go to previous page in this tab
Session
