Prediction of Daily Suspended Sediment Load Using Radial Basis Function Neural Networks
Abstract
In this study, daily suspended sediment amount were predicted from corresponding daily streamflow by using Artificial Neural Networks (ANNs). Radial Basis Functions (RBFs) were chosen as ANN method and two different learning algorithms were applied namely Quickprop (QP) and Delta-bar-Delta (DBD) with two different transfer functions called linear tangent hyperbolic axon (litanhaxon) and tangent hyperbolic axon (tanhaxon). Prediction was made by using flow and suspended sediment data of Ispir gauging station on Çoruh River, Turkey between 1991 and 1999. The data, 106 in total, were used as calibration/training and validation/testing sets for the chosen RBF neural network architecture. Of the data obtained 76 measurements (72%) were reserved for the calibration and the remaining data were used for validation. All developed RBF networks have one hidden layer (HL) and one process element (PE) or neuron. Mean Absolute Error (MAE) and coefficient of correlation (R) were used as performance criteria. According to MAE performance criteria of developed networks, DBD learning algorithm with litanhaxon (MAE=0.052) gave best results and following QP learning algorithm with litanhaxon (MAE=0.054), DBD learning algorithm with tanhaxon (MAE=0.056), QP learning algorithm with tanhaxon (MAE=0.057), respectively. According to R performance criteria, DBD learning algorithm with tanhaxon gave best results (R = 0.963) and following QP learning algorithm with tanhaxon (R=0.961), QP learning algorithm with litanhaxon (R=0.955), DBD learning algorithm with litanhaxon (R=0.945), respectively. This study showed that RBF Networks provide satisfactory results in engineering applications for prediction of suspended sediment amount from corresponding daily streamflow by using ANN.
Keywords
Kaynakça
- Aik, LE and Zainuddin, Z 2008.An Improved Fast Training Algorithm for RBF NetworksUsing Symmetry-Based Fuzzy C-Means Clustering.MATEMATIKA, 24(2), 141–148.
- Bors, A, G 2001. Introduction of the Radial Basis Function (RBF) Networks, Online Symposium for Electronics Engineers, issue 1, vol. 1, DSP Algorithms: Multimedia, pp. 1-7.
- Caroni, E, Singh,V, Pand Ubertini, L 1984. Rainfall-runoff-sediment yield relation by stochastic modeling. Hydrological Sciences Journal, 29:2, 203-218
- Cigizoglu, HK 2002. Suspended Sediment Estimation and Forecasting using Artificial Neural Networks.Turkish J. Eng. Env. Sci.26: 15-25.
- Fahlman, SE 1988.Fast-learning variations on back propagation: an empirical study. In Proceedings of the1988 Connectionist Models Summer School, D.S.
- Fashtali, FJ 2003. Landuse change and Suspended Sediment Yield Analysis Using RS and GIS a case study in Uromieh lake area (Shar-chi Catchment). International Institute For Geo-Information Science and Earth Observation Enschede, The Netherland.
- Haykin, S 2001. Neural Networks: A comprehensive foundation. Macmillan College Publishing.
- Kerr, SJ 1995. Silt, turbidity and suspended sediments in the aquatic environment: an annotated bibliography and literature review. Ontario Ministry of Natural Resources, Southern Region Science & Technology Transfer Unit Technical Report TR-008. 277 pp.
- Kisi, O, Yuksel, I and Dogan E 2008.Modelling daily suspended sediment of rivers in Turkey using several data-driven techniques. Hydrological Sciences Journal, 53:6, 1270-1285
- Knighton, D1998. Fluvial Forms and Processes: A New Perspective, Arnold, London.