우수관망 유출량 예측을 위한 시공간 딥러닝 모델 개발 = Development of a spatio-temporal deep learning model for urban drainage outflow prediction

다국어 초록 (Multilingual Abstract)

This study developed a spatio-temporal deep learning-based model for accurate prediction of stormwater outflow in urban drainage systems. In order to reflect temporal and spatial characteristics affecting runoff simultaneously, CNN-RNN model was designed by combining Convolutional Neural Networks (CNNs) which extract spatial features, with Recurrent Neural Networks (RNNs), especially Long-Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) models, which handle temporal sequences. The proposed model was applied two urban drainage basins in Seoul : Gasan 1 rainwater pump station basin and Hagye basin. Its performance was evaluated by comparing the synthetically generated data from an urban drainage network model, using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Nash-Sutcliffe Efficiency (NSE) as evaluation metrics. In the results, CNN-LSTM and CNN-GRU models outperformed the respective single models (CNN, RNN) demonstrating lower prediction errors. These findings suggest that stormwater runoff in urban drainage systems is influenced by both spatial and temporal factors, and support the possibility of utilizing spatiotemporal deep learning-based models. Furthermore, the model is expected to be applicable for constructing real-time prediction systems that utilize smart sensor data, maintaining high prediction accuracy even with limited observational data.