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

Area of Science:

• Environmental Engineering
• Urban Hydrology
• Sustainable Urban Development

Background:

• Urbanization exacerbates waterlogging issues during rainfall.
• Rain gardens are crucial for managing urban surface runoff and mitigating waterlogging.
• Understanding factors influencing rain garden effectiveness is vital for urban planning.

Purpose of the Study:

• To predict the runoff control effectiveness of rain gardens using multiple modeling approaches.
• To identify key parameters influencing rain garden performance.
• To optimize predictive models for enhanced accuracy.

Main Methods:

• Collected 240 runoff control rate datasets from five experimental rain garden structures.
• Performed feature correlation analysis to identify critical input parameters (rainfall recurrence interval, storage layer depth, catchment area, infiltration rate).
• Developed and optimized predictive models (BP, SVM, Random Forest) using the Zebra Optimization Algorithm (ZOA).

Main Results:

• The Zebra Optimization Algorithm-optimized Backpropagation (ZOA-BP) model demonstrated superior prediction accuracy (R² = 0.979).
• The ZOA-BP model achieved a Root Mean Squared Error (RMSE) of 2.331, validating its predictive capability.
• Key parameters influencing runoff control were identified and incorporated into the models.

Conclusions:

• The ZOA-BP model provides a reliable tool for predicting rain garden runoff control effectiveness.
• This research offers valuable insights for designing and implementing effective rain garden systems.
• The findings can help reduce design and operational costs for urban water management projects.