 This paper proposes a deep learning technique-based data-driven model for urban flood prediction. It integrates LSTM networks, Bayesian optimization, and transfer learning techniques to achieve fast and accurate predictions. The model was tested in a case study in northern China and demonstrated its accuracy and efficiency. Compared to other models, it has shown superior performance in terms of speed and accuracy. Furthermore, the model's transferability was also validated by applying it to a new case study. This article was authored by Q-Show, S-Tang, Z-Situ, and others.