Urban flood prediction based on PCSWMM and stacking integrated learning model

With global warming and urbanization accelerating, urban flood disasters have become increasingly frequent, highlighting the need for reliable urban flood forecasting models. Traditional numerical simulation models and individual machine learning model often suffer from poor robustness and low efficiency, leading to inaccurate predictions. Meanwhile, current machine learning models have high requirements for sample size and quality of training data, which can be challenging to meet even with data interpolation. To address these limitations, this study proposes an urban flood forecasting method that combines the strengths of PCSWMM numerical simulation and stacking ensemble learning. The key objectives of this research are to: (1) Leverage the reliable data and features generated by the PCSWMM numerical simulation to train a robust machine learning model; (2) Employ the stacking ensemble learning algorithm to integrate multiple base learner models, thereby reducing the errors caused by individual model deficiencies and improving the overall prediction accuracy and stability. The results demonstrate that the data obtained through numerical simulation has stable predictive ability and can provide reliable datasets and features for training machine learning models. Compared with KNN, XGBoost, and LightGBM models, the stacking ensemble model has the highest accuracy, with RMSE improvements of 69.92%, 51.82% and 73.79% respectively. This indicates that the stacking ensemble learning method is superior to the individual machine learning model, reducing prediction errors and improving overall prediction performance. The findings of this study offer a new perspective for urban flood forecasting and provide a reliable basis for flood disaster simulation, contributing to the field of urban hydrology and disaster risk management.