Smart Operation of Dams with AI
Smart Operation of Dams with AI
  • Reporter Kim Yu-jin
  • 승인 2023.12.05 20:42
  • 댓글 0
이 기사를 공유합니다

▲Prof. Jonghun Kam, Eunmi Lee (integrative course) and Water Level Prediction Graph by GRU Model (from left)
▲Prof. Jonghun Kam, Eunmi Lee (integrative course) and Water Level Prediction Graph by GRU Model (from left)

  In Aug. 2020, heavy rainfall and a substantial water release in the Seomjin River Dam caused a tremendous flood, causing 160 billion KRW in damage. One of the primary issues that worsened the damage was the failure in the dam operation, as the operators did not discharge enough water before the heavy rainfall.
  Could this have been averted if we knew in advance and took measures? Research led by Professor Jonghun Kam (DESE) and Eunmi Lee (integrative MSE/PhD program in DESE) in the Hydro-Climatology Research (HCR) group, explored the efficient operation of dams using deep learning and verified its effectiveness. The study was published in the Journal of Hydrology.
  Dams are responsible for managing and mitigating water disaster such as floods and droughts in a timely manner. But with the rapid climate change, unexpected floods and droughts make dam operation more challenging. Scientists attempted to use artificial intelligence (AI) models that learn from water and climate-related big data.
  The HCR group focused on predicting the operation pattern of dams in the Seomjin River area using an AI model and developing an explainable scenario on how the trained AI model predicts the water level. They used a deep learning algorithm, the Gated Recurrent Unit (GRU) model, trained by the data from dams in Seomjin River over 2002 to 2021. They used rainfall, inflow, and outflow data as inputs and the dam’s hourly water level as output. The GRU model showed high accuracy of over 0.9.
  Secondly, they designed explainable scenarios by altering input values and investigated how the trained GRU model determined the water level. Changes in inflow rates had a significant impact to the simulated water level, while rainfall did not. The GRU model also learned the specific-site rule of dam operation and designed different operation patterns for each dam.
  Prof. Kam said, “We not only analyzed dam operation methods and patterns using AI and verified its effectiveness, but also presented a methodology to indirectly understand how AI models known as Black Box Models determine water levels of a dam reservoir.” “We hope this study will help understand the operation rule of dams and improve it more efficiently in the future,” he said. 
  The research was supported by the National Research Foundation of Korea (NRF).