Neural Network Models for Captured Runoff Prediction of Permeable Interlocking Concrete Pavements

Abstract

Impervious area expansion has substantially decreased infiltration into the ground, increased flood risk and carried contaminant materials into surface water. Permeable pavement is one type of Low Impact Development (LID) practices which can capture surface runoff water during storm events and keep the combined sewer system from overflowing. Developing a prediction model to estimate the captured runoff volume from watershed area by permeable pavements provides valuable information. A new model has been derived to more accurately predict the captured surface runoff volume using Artificial Neural Networks (ANNs). The proposed model relates rainfall parameters and site characteristics to the runoff volume captured by the permeable pavements. The database used for developing the prediction models is obtained from the collected data of the monitored permeable pavement. A parametric study is completed to determine the sensitivity of the effective parameters on the captured runoff volume. The results indicate that the proposed model is efficiently capable of estimating the captured runoff by the permeable pavements for different rain events and site characteristics. The ANN model considers all the contributing factors and provides more precise volume prediction than the linear model. This information can be used to schedule more effective maintenance treatments and improve Permeable Interlocking Concrete Pavement (PICP) design.