AI모형과 강우-수위 자료의 교차상관분석을 적용한 도심지 하천 홍수위 예측

초록

Extreme rainfall events associated with climate change have increased flood risks in urban streams, potentially causing road inundation and flooding in underground spaces. This study developed AI-based models for predicting flood-season water levels using rainfall and water level data and evaluated whether considering the rainfall–water level lag improves prediction performance. Cross-correlation analysis was conducted to determine the rainfall lag most strongly related to the current water level, and the identified lagged rainfall data were used as model inputs. LSTM and Bi-LSTM models were then developed and evaluated using statistical performance metrics. Rainfall from two hours earlier showed the strongest correlation with water levels. During the 2023 validation period, the LSTM and Bi-LSTM achieved NSE values of 0.72 and 0.75, respectively, while both models achieved an NSE of 0.69 in 2024. These results demonstrate the applicability of lag-adjusted AI models to urban stream flood-level prediction and flood response in underground spaces.

키워드