Prediction of flow discharge in compound open channels using adaptive neuro fuzzy inference system method

Abstract

Discharge estimation in rivers is the most important parameter in flood management. Predicting discharge in the compound open channel by analytical approach leads to solving a system of complex nonlinear equations. In many complex mathematical problems that lead to solving complex problems, an artificial intelligence model could be used. In this study, the adaptive neuro fuzzy inference system (ANFIS) is used for modeling and predicting of flow discharge in the compound open channel. Comparison of results showed that the divided channel method with horizontal division lines with the Coefficient of determination (0.76) and root mean square error (0.162) is accurate among the analytical approaches. The ANFIS model with the coefficient of determination (0.98) and root mean square error (0.029) for the testing stage has suitable performance for predicting the discharge of flow in the compound open channel. During the development of the ANFIS model, found that the relative depth, ratio of hydraulics radius and ratio of the area are the most influencing parameters in discharge prediction by the ANFIS model.

Introduction

Flood is one of the most dangerous natural phenomena, which usually occur in the river consisting of a main channel flanked by one or two floodplains (Fig. 1). Studies on the hydraulic of rivers, specifically at the unsteady flow conditions are the major part of the hydraulic engineering researches which are named flood engineering [10]. Flood engineering includes two main concepts: hydrological and hydraulics fields. Hydrological field studies focus on topics such as the hydrograph of flood and in the hydraulics field studied on subjects such as flood routing, calculating the water surface profile, velocity distribution and sediment transport [8], [9], [24]. Calculating the discharge and water surface profile in rivers are the main common topics that are usually discussed in the flood engineering. Rivers hydraulic is more complex especially when the flood is occurring because the flow structure is fully turbulent and 3D dimensionality [7]. Although for calculating the discharge in rivers, conventional formulas such as Manning and Chezy formulas have been proposed, but to achieve greater accuracy in estimation, these formulas should be modified. Several ways such as analytical approaches have been proposed for this purpose and in this regard the compound open channel concept is punctual [24], [41]. The normal river discharge flows in the main channel but when the flood happens, the water level begin to increase therefore, additional discharge flows in the flood plain area [6]. The floodplains are usually covered by vegetation so these are rougher than the main channel [12]. The difference between the roughness of the main channel and the floodplains leads to different velocity in the cross section of the flow. This difference in velocities, shown in the Fig. 1, creates an interactive area which includes the vortices flow and causes of the momentum transferring between the main channel and the floodplain. The main difference between the traditional channel concept and compound open channel is related to this area [19], [36]. Calculating the capacity of the compound open channel by classical formulas leads to incorrect estimations of the discharge flow in comparison with real measured data [2], [3], [13], [17]. Recently, analytical approaches have been proposed to increase the accuracy of the discharge calculation. In this regard, the single-channel method (SCM), divided-channel method (DCM), and the coherence method (COH) can be stated [43], [44]. Seckin [31] applied the SCM, DCM and COH methods to calculate the discharge of flow in compound channel and showed that the COH method is more accurate among the other analytical approaches. Seckin et al. [32] implemented the quasi 2-D dimensional Shiono and Knight Method (SKM) for calculating the flow discharge in compound channels and in the follow; they compared the SKM and 1D traditional method such as SCM and DCM. The results of these researches showed that SKM has a better performance in comparison with the traditional method. Mohanty and Khatua [18] proposed a modified divided-channel method (MDCM) to calculating discharge in compound channel and stated that the result of this method is suitable for practical problems. Recently, by advancing the soft computing techniques in engineering, researchers turned to use these methods for modeling the hydraulic phenomena. In this regard, Sahu et al. [30] and Unal et al. [39] used the artificial neural network (ANN) model to predict the flow discharge in the open compound channel. Azamathulla and Zahiri [5] and Zahiri and Azamathulla [46] applied the Genetic programming (GP) model to predict the flow discharge in the compound channel. Parsaie et al. [28] applied the support vector machine (SVM) technique for prediction of flow in the compound open channel. In this study, an evaluation is conducted on the performances of analytical approaches and then the ANFIS is developed for modeling and prediction of the flow discharge in the compound open channel. Based on the reports, this method is more reliable [29].