Prediction of breach peak outflow and failure time using artificial neural network approach

Prediction of breach peak outflow and time of failure with appreciable level of accuracy is of substantial importance to avoiding potential life loss, minimising damage and consequently financial losses in the downstream floodplain. The damage is certain when a dam fails; however, the magnitude of it cannot be evaluated a head of time. This paper proposes the use of Artificial Neural Network (ANN) approach to predict the peak outflow and failure time of breached earthen dams. Several parameters such as the type of dam, height and volume of water behind the dam, erodibility of dam materials, and the mode of failure are used for the estimation purpose. Historical datasets of dam failures are employed in the training process of various ANN structures. The reliability of the proposed ANN approach was evaluated by means of Correlation coefficient (CC) and the Root Mean Square Error (RMSE). Subsequently, a comparison is drawn between ANN approach and popular regression models. The ANN approach is found to be considerably more reliable than regression analysis.