Prediction of rain garden runoff control effects based on multiple machine learning techniques.

Due to the rapid development of urbanisation, cities frequently experience waterlogging during rainfall. Rain gardens are widely used in new urban construction because they effectively control surface runoff from rainwater, thereby reducing waterlogging. The runoff control effectiveness of rain gardens is influenced by multiple factors. This paper predicts the runoff effects of rain gardens using multiple models. By constructing five experimental structures, 240 sets of runoff control rates for rain garden structures were collected to build a database. Feature correlation analysis identified four input parameters: rainfall recurrence interval, storage layer depth, catchment area, and infiltration rate. Using BP, SVM, and Random Forest, initial predictive models for the runoff control effectiveness of rain gardens were established. To enhance the accuracy of the models, the Zebra Optimization Algorithm was employed for optimisation, and model performance was characterised using the coefficient of determination, mean squared error, and mean absolute error. The results show that the ZOA-BP model has the best prediction results on the test set, the prediction accuracy (R²) is 0.979, and the RMSE is 2.331, which verifies the validity of the model. This research outcome can provide references for the application of rain gardens and is expected to reduce the design and operational costs of related projects.