Comparing three deep learning models for prediction of water levels in the urban drainage systems

This study investigates the application of deep learning models, including LSTM, TCN, and TFT, for urban water level forecasting using synthetic time series data. The models exhibited high predictive accuracy, with two achieving an R² greater than 0.99, and reliably reproduced observed system dynamics. These results indicate that advanced deep learning approaches can provide accurate forecasts in urban hydrological contexts and offer a basis for further development of data-driven flood management strategies.