머신러닝 모형을 이용한 기후변화에 따른 충청권역 생활용수 수요량 예측

초록

The importance of predicting water demand for stable water resource management is growing as water use increases due to economic growth and climate change. In Korea, the National Water Management Master Plan (NWMMP) predicted domestic water demand in every five years from 2020 to 2030. However, social, economic, and meteorological factors were not considered in the NWMMP. In this study, using XGBoost, DNN, and LSTM, we predicted domestic water demand considering social, economic, and meteorological influencing factors provided in climate change scenarios. The LSTM model outperformed other models, achieving an average R2 of 0.98 in the training phase and 0.94 in the validation phase. Accordingly, LSTM was selected to achieve high accuracy and to avoid low risk of overfitting. Three distinct patterns of predicted domestic water demand emerged: decreasing, stabilizing, and increasing. and these patterns were evident in the representative regions of Gyerong, Asan, and Jeungpyeong, respectively. Compared to the domestic water demands derived solely from the planned water supply population in the NWMMP, the increasing trend was similar. The LSTM predicted increases of 0.44%(SSP1), 0.87%(SSP2), and 1.32% (SSP5) by 2035 relative to 2020, while the NWMMP predicted increases of 14.33%(high demand), 8.00%(reference demand), and 16.81%(low demand). These results are expected to support the development of regionally customized water policies.

키워드