AI-powered dams optimize water and electricity
Newswise — In August 2020, after a period of prolonged drought and intense rains, a dam near the Seomjin River in Korea overflowed during a water release, resulting in damages exceeding 100 billion won (76 millions of dollars). The flooding was attributed to the dam’s water level remaining 6 meters above standard. Could this incident have been avoided thanks to predictive management of dams?
A research team led by Professor Jonghun Kam and Eunmi Lee, a doctoral student from the Division of Environmental Science and Engineering at Pohang University of Science and Technology (POSTECH), recently used deep learning to examine dam operating models and evaluate their effectiveness. Their findings were published in the Journal of Hydrology.
Korea faces peak rainfall during the summer and relies on dams and associated infrastructure for water management. However, the escalating global climate crisis has led to the emergence of unforeseen typhoons and droughts, complicating the operation of dams. In response, a new study has emerged that aims to outperform conventional physics models by harnessing the potential of an artificial intelligence (AI) model trained on vast, big data.
The team focused on building an AI model aimed at not only predicting the operational patterns of dams in the Seomjin River basin, specifically focusing on the Seomjin River Dam, the dam of Juam and the Juam check dam, but also to understand the decision-making processes of the trained AI models. Their goal was to formulate a scenario describing the methodology for forecasting dam water levels. Using the Gated Recurrent Unit (GRU) model, a deep learning algorithm, the team trained it using data spanning 2002 to 2021 from dams along the Seomjin River. Data on precipitation, inflow, and outflow were used as inputs while hourly dam levels were used as outputs. The analysis demonstrated remarkable accuracy, with an efficiency index greater than 0.9.
Subsequently, the team designed explainable scenarios, manipulating the -40%, -20%, +20%, and 40% inputs of each input variable to examine how the trained GRU model responded to these changes in values. entries. Although changes in precipitation had a negligible impact on dam water levels, variations in inflow significantly influenced dam water levels. Notably, the identical change in outflow yielded different water levels in distinct dams, affirming that the GRU model had effectively learned the unique operational nuances of each dam.
Professor Jonghun Kam remarked: “Our review went beyond predicting dam operating patterns and secured their effectiveness using AI models. We introduced a methodology aimed at indirectly understanding the decision-making process of the AI-based black box model determining dam water levels. He added: “Our aspiration is that this information will contribute to a deeper understanding of dam operations and improve their efficiency in the future. »
The research was sponsored by the Mid-Career Research Program of the National Research Foundation of Korea.
