Comparison of the Artificial Neural Networks and the Linear and Multiple Regression Techniques in Predicting the Rate and Distribution of Sedimentation in Man-Made Reservoirs

Article abstract:Erosion and transport of suspended and bed loads are of the most complex and important problems of hydrodynamics inherent in the study of water supply projects. As the artificial neural networks (ANRs) are endowed with two fundamental characteristics of learning based on empirical data (the potential and ability to generalize) and parallel structuralism, they are one of the most important versions of artificial intelligence in which, by drawing inspiration from human brain, the desired data are saved in the networks. In this study we not only designed the ANRs applying the error back scattering method and stability analysis, and the convergence of control parameters of its ring systems, but also implemented it to assess its performance in predicting the rate and distribution of sediments in the Ekbatan Reservoir. Moreover, we employed the linear and multiple regression techniques for the same purpose and compared the results with those of the ANRs. The regression coefficients RSME and R2 indicated that the ANRs are the most accurate method to predict the rate and distribution of sedimentation in a reservoir. Furthermore, the linear regression is preferred over the multiple regression due to the mechanism of error distribution in non-linear calculations