Application of Artificial Neural Network for Predicting Peak Discharge from Breached Embankment Dam

Abstract

Estimation of peak discharge is a key parameter for risk assessment in case of dam failure, and has attracted great attention from researchers in recent years. Many formulas are available in the literature, but these cannot cover all experimental scenarios. Existing models are typically inadequate to address the complexities of dam breaches. This research attempted to predict the peak discharge in the breached embankments with an artificial neural network (ANN) model, which is effective in nonlinear problems, using datasets obtained from various dam breaches cited in the literature. The ANN model is useful in the preparation of emergency action plans since it enables prediction of peak discharge. Multilayer Perceptron (MLP) with Levenberg-Marquardt (LM) and Bayesian Regularization (BR) algorithms was used to predict peak discharges from breached embankments. The dataset was divided into three: 56% for training, 20% for validation and 24% for testing. Different scenarios were created using different input combinations. Performance evaluation was based on the root-mean squared error (RMSE), percent bias (PBIAS), determination of coefficient (R2), Nash-Sutcliffe efficiency (NSE) and RMSE-observations standard deviation ratio (RSR). A comparison of training algorithms revealed that LM showed the best performance when the best ANN was selected from 1000 networks. Volume of water above the breach bottom (Vw) had a greater effect on model performance than the depth of water above the breach bottom (Hw). The best performance was obtained when both Vw and Hw were used as input. © 2024 ISHS. All Rights Reserved.