Estimation of Velocity Field in Narrow Open Channels by a Hybrid Metaheuristic ANFIS Network

Abstract

Owing to the significance of velocity distribution, several laboratory and simulation studies were done on the velocity distribution in open channels. Laboratory and field studies show that the maximum value of velocity through narrow open flumes takes place beneath the water surface knowing as the velocity dip mechanism. This velocity dip is a highly complex feature of flow in narrow channels, and no expression can evaluate it exactly. This study introduces a hybrid Metaheuristic model to estimate the velocity distribution within the narrow open canals. To optimize the linear and nonlinear parameters of adaptive neuro-fuzzy inference systems (ANFIS) models, singular decomposition value (SVD) and genetic algorithm (GA) are employed. In order to increase the flexibility of the model for an optimal design, two different objective functions are used, and the superior optimal point using the Pareto curve is estimated. To evaluate the accuracy of the hybrid ANFIS-GA/SVD model, the velocity distribution in three hydraulic circumstances is compared to the measured values. ANFIS-GA/SVD predicts velocity distribution with reasonable accuracy and estimates the velocity dip value with high precision. The Root Mean Squared Error (RMSE) for depths, D = 0.65 m, D = 0.91 m, and D = 1.19 m is calculated 0.052, 0.044, and 0.053, respectively. According to the numerical model results, the ANFIS-GA/SVD simulated the velocity distribution in depth of D = 0.91 m more accurately than other ones. Also, almost 94%, 96% and 88% predicted velocities for D = 0.65 m, D = 0.91 m, and D = 1.19 m, respectively that are modeled by the ANFIS-GA/SVD algorithm show an error of less than 10% for all measured data in the entire cross-section.