Please use this identifier to cite or link to this item: http://localhost:8081/xmlui/handle/123456789/11901
Title: RAINFALL - RUNOFF MODELING USING ARTIFICIAL NEURAL NETWORK FOR MOHEGAON SUB - BASIN
Authors: Goswami, Sudhir
Keywords: HYDROLOGY;RAINFALL - RUNOFF MODELING;ARTIFICIAL NEURAL NETWORK;MOHEGAON SUB - BASIN
Issue Date: 2001
Abstract: Artificial neural network (ANN) was developed in the early 40's with a view to emulate the neuro-biological system of human• brain. Now a days this modeling technique is widely being used in the field of hydrology and other branches of water resources engineering problems. Among the learning procedure of ANN, Rumelhart's(1986) back propagation (BP) learning scheme is. a remarkable development in this field and has widely been used by the various investigators. While applying this scheme to ANN, the researchers have found that the convergence procedure included in this scheme is very slow and near the minimum it suffers from significant amount of oscillations(E.M. Johansson, F.U. Dowla and D.M. Goodman, 1992) causing the scheme impractical for a very large network structure of ANN. Conjugate gradient optimization is an efficient tool and technique that ascertains the location of the optimum in a fixed number of iterations. Fletcher and Revees (1964) developed a technique for this optimization procedure that uses a set of direction of search , called conjugate directions which are obtained by the knowledge of the gradients of the current and previous point of search only. This method improves on the optimization procedure and time required for training of the ANN model. In the present study, a conjugate gradient algorithm has been developed and a subroutine in FORTRAN language has been written for this optimization technique as suggested by Fletcher and Revees(1964). Adding this subroutine to artificial neural network, a conjugate gradient artificial neural network has been developed. The developed model has been applied to calibrate, cross-validate and verify the rainfall runoff process of the Mohegaon sub-basin of the upper Narmada basin. A total of nine years data (from 1981-82 to 1989-90) have been used for this purpose . Three years data are used for calibration of the model, another three years data for cross-validation and rest three years data for verification of the model. The monthly, weekly and daily M rainfall and runoff data for the Mohegaon sub-basin of the upper Narmada basin have been used for this study.
URI: http://hdl.handle.net/123456789/11901
Other Identifiers: M.Tech
Research Supervisor/ Guide: Agarwal, Avinash
Singh, R. D.
Singh, Ranvir
metadata.dc.type: M.Tech Dessertation
Appears in Collections:MASTERS' THESES (Hydrology)

Files in This Item:
File Description SizeFormat 
HYDG10830.pdf3.04 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.