Artificial neural networks for estimating the hydraulic performance of labyrinth-channel emitters

Abstract

In this paper, we examine the discharge of labyrinth-channel emitters under different operating pressures (P) and water temperatures (T). An artificial neural network (ANN) and multiple linear regression (MLR) model are developed for the emitter flow variation (qvar) and the manufacturer’s coefficient of variation (CV). As well as P and T, the structural parameters of the labyrinth emitter are considered as independent variables. The ANN results demonstrate that a feed-forward back-propagation network with five input neurons and 14 neurons in the hidden layer successfully model qvar and CV. The trapezoidal unit spacing and path length of the labyrinth emitter are found to be insignificant. In our ANN model, we use a hyperbolic tangent as the activation function in the hidden layer and the output layer. Statistical criteria indicate that the ANN is better at predicting the hydraulic performance of the labyrinth emitters than MLR. The root mean square errors for qvar and CV are 1.0497 and 0.0044, respectively, for the ANN model, and 2.0703 and 0.0107, respectively, for the MLR model using a test dataset. The relatively low errors obtained by the ANN approach lead to high model predictability and are feasible for modeling the hydraulic performance of labyrinth emitters.