Scour Depth Prediction in Sand Beds using Artificial Neural Networks and ANFIS Methods

Abstract

Background/Objectives: In this study, the maximum scour depth modeling of Multi-Layer Perceptron based artificial neural network (MLP), Radial Basis Function based neural network (RBF), Adaptive Neural-Fuzzy Inference System (ANFIS), multiple regression analysis and empirical relationships of Ingis, Shen & Schneider, Laursen & Toch, Breusers, Jain, Melville & Sutherland, and Melville & Chiew were used. Methods/Statistical Analysis: To compare the performance of each method, the standard tests such as standard Root Mean Square Error (NRMSE), standard Mean Absolute Error (NMAE), correlation coefficient (R) and Nash Sutcliffe index were implemented. Findings: The results demonstrate that the adaptive neural-fuzzy inference system with the lowest standard root mean square error (0.665) and the highest correlation coefficient (0.76), has the best performance compared to other methods Melville and Sutherland experimental method compared to other empirical methods yields more suitable performance (NRMSE = 1.058). The standard root mean square error of perceptron neural network, radial basis neural networks and multiple regression analysis were calculated as 0.665, 0.94 and 0.95 respectively. Application/Improvements: Percent improvement in the performance of fuzzy neural networks, perceptron neural networks, radial basis neural networks and multiple regression analysis compared to Melville and Sutherland 37.14, 27.60, 11.15 and 10.21 respectively.