Skip to contentSkip to footer

Ja-Ho Koo joined IHE Delft as a PhD student in December 2020. Before starting a PhD journey, he has worked in the Korea Water Resources Public Corporation (K-water) since 2004. (He is receiving a scholarship from K-water to study at IHE Delft.) He worked as a reservoir operator of the Han and Geum river basin of Korea as well as a supervisor of the construction of the Siwha tidal power plant, which is the largest tidal power plant in the world. Recently, he have worked in a newly built department which organized to introduce the emerging AI and data technologies to the water sector. In this department, he took part in many projects relating to the database, machine learning and deep learning. Therefore, database, DL and GIS programming using Python is also one of the areas he is interested in.

In 2015, he got a master degree in public administration in Korea. It was not much difficult for him to come up because he had had 2 bachelor degrees, which are civil engineering and business administration. He wrote a master thesis using a game theory and hydraulic/hydrology simulation models. It was a really enjoyable experience to try to combine knowledge of 2 different parts. 

Nowadays, his interest has changed into reinforcement learning as well as dynamic programming and function approximation. It is new and not easy for him, but he want to know about it more deeply. Why? Because it is fun!

Research Summary

Reservoirs operation is highly dependent on the quality of rainfall forecast, and all kinds of weather forecasts can only be uncertain especially in heavy rainfall. Therefore, to ensure the safety of dams and prevent/reduce damages to downstream areas, a model generating robust reservoir operation policies under meteorological uncertainties is really necessary. To develop this model, physically-based or data-driven models will be built at first, and used to generate expected hydrological outputs such as reservoir inflow and water levels of rivers. A probability model which replicates uncertainties of rainfall forecast is also needed. Then optimization algorithms can be applied to find the best policies based on the results of the previous models. Clearly, the policies proposed by this model should be presented hourly and pursue robustness against floods. Finally, this model will be verified by applying it to the Geum river basin in Korea using historical data.