APPLICATION OF ARTIFICIAL NUERAL NETWORK FOR ESTIMATION OF COEFFICIENT OF DISCHARGE FOR SIDE COMPOUND WEIR

Abstract

Side weirs are the most common diversion structures used for flow control, irrigation, drainage networks, and waste-water channels. In this study an alternative approach to the conventional regression approach in the form of artificial neural network is proposed to predict the coefficient of discharge (Cd). The performance of artificial neural networks over regression approach is assessed here. The experimental data collected in the present investigation having wide range of hydraulic and geometrical variable are used for training and validation of the ANN network. It was found that coefficient of discharge of a compound side weir is a function of upstream Froude number (F1), ratio of weighted crest height of weir to crest length of side weir and ratio of upstream depth of flow in channel to crest length of side weir (Y1/L). A network architecture complete with trained values of connection weight and bias and requiring input of grouped parameters pertaining to , Y1/L, F1 & B/L is recommended in order to predict the coefficient of discharge. On the basis of F-test, it is observed that is the most significant parameter for the prediction of coefficient of discharge. But in view of the variability in the outcome resulting from the application of different ANN models, it is felt that network which require all input quantities, may be followed for generality. The ANN, (FFBP) Model, has the highest R (0.871) lowest MAPE (15.674) and RMSE (0.0699).