머신러닝을 활용한 대청수계 취수원 냄새물질 예측

The contamination of source water by odor compounds are one of the problems related to the water quality management, especially in Daecheong Reservoir, South Korea issued an algal alert system anually. This study investigated the efficiencies of 4 machine learning models, including Multi-parameter Regression Analysis(MRA), Decision Tree(DT), Artificial Neural Network (ANN) and Random Forest (RF), for odor compounds forecasting(Geosmin, 2-MIB) in the Daecheong Water Intake Station, where supply water treatment plants to source water. The models based on input variables considered correlation between target output and water quality parameter and hydrologic·meteorological factors. The established models showed good results between observed and simulated values. For Geosmin models, ANN produced better forcasting results than others. RF showed the best results for 2-MIB models. These results and models are applied in work-site operations through the Daecheng Intergreted Water Quality Forecasting System since September, 2018.