Please use this identifier to cite or link to this item:
|Title:||NEURO-FUZZY MODEL APPLICATION TO SEDIMENT DISCHARGE IN NARMADA RIVER|
|Abstract:||Rivers are the most important geological agents on the surface of the continents and are the major pathways for transfer of continental materials to the oceans, in the form of water, dissolved and suspended loads. River transport of materials has been considered for over a 100 years in geological budgets. It gives essential information, on both processes affecting the continental surface and on the amount and nature of materials carried to water bodies. Sediment transfer from continents to oceans via rivers, is one of the most important processes regulating river-bank stabilization, soil formation, and many other earth-related processes. Due to the changes in continental positions during the geologic past, water flow and sediment loads in rivers, have also shown variations during different time periods. Factors such as relief, channel slope, basin size, seasonality of rains and water discharge mainly controls sediment loads in rivers. Human interventions in the form of reservoirs for water storage have impounded and trapped huge sediment loads on the continental parts. The Narmada River flows through the Deccan volcanics and transports water and sediments to the adjacent Arabian Sea. The average annual sediment flux to the Arabian Sea is 34.29 million tons (0.2% of global sediment flux) and water discharge of 23.57 km3 annually. Measurements of sediment loads along the river stretch at selected locations, are being carried out by various government organizations, such as, Central Water Commission, Narmada Basin Organisation and others. In the present thesis, an attempt has been made to correlate variability of sediment loads with rainfall and water flow and to predict sediment load based onartificial intelligence. The model application has been very robust and may further be extended to other rivers within extremely high accuracy. Sediment discharge is predicted using two models, neural networks and hybrid fuzzy neural network. The methodology comprised of collection of data of rainfall, water discharge and sediment discharge of Narmada river at various locations and applied to develop artificial neural networks and hybrid fuzzy neural networks for the prediction of futuristic sediment discharges. In the proposed method, a three layer artificial neural network (ANN) model based on back propagation algorithm has been used for analysis. In the proposed ANN model day, month, year, rainfall and water discharge are taken as the input parameters which mainly influence the sediment discharge as output parameter. A total of 523 (out of 549) patterns are used for training the ANN and those patterns are for three years (1997-99) monsoon (June-Nov.) period. The proposed ANNmodel adopts feed forward back propagation algorithm with 5 input neurons (represent day, month, year, rainfall and water discharge) in the input layer, 70 neurons in the hidden layer and one neuron (represent sediment discharge) in the output layer. Hence, the proposed model has been trained with hidden neurons varying between 1 to 100 and it has been observed that hidden layers with 70 hidden neurons converged with minimum error against fixed number of 10000 iterations. Initially very small positive as well as negative weights are assigned to the model and were upgraded in the end of every iteration. The process is continued till the minimum mean absolute error (10"9) is achieved or end iteration (10000) is reached. During fuzzy-neural network design, both crisp and linguistic values are used to represent input layer. Day, month and year are treated as crisp variables, whereas water discharge and rainfall are treated as fuzzy variables. Triangular and Trapezoidal membership functions are used to represent low, medium and high range of both rainfall and water discharge magnitudes. In the proposed fuzzy neural network model, variations in both rainfall and water discharges are being represented either low, medium or high category i.e. each fuzzy variable is represented bythree neurons. Execution time and errors are expected to reduce in case of fuzzy-neural network training compared to neural network, mainly due to reduction in non-linearity. For fuzzy neural networks, the number of iterations (epochs) and minimum error for convergence are assumed same as ANN. After training, the system is simulated with the same input neurons, which are used in training as the test data randomly to validate the performance ofthe proposed model. The corresponding sediment load output from the output layer for test patterns are captured and compared with the targeted values. To validate the performance of the proposed models, both ANN and fuzzy neural network models are applied to 11 important sites covering the entire Narmada river. To maintain uniformity in the results, the simulation parameters and stopping criteria are taken same for all the study locations and for both models. In general it has been observed that the fuzzy-neural network is more robustcompared to ANN. The thesis has been organized into six chapters. Chapter 1: A brief introduction to artificial neural networks and hybrid fuzzy neural network application to river sediment discharge has been presented in this chapter. Athorough literature review on river studies is being discussed, which includes mainly the determination of river parameters using artificial neural network and fuzzy models. Chapter 2: In this chapter, various features (water discharge, rainfall, load sediment discharge, etc.) of the study area i.e., "The Narmada River Basin" are presented in detail. Chapter 3: This chapter presents a brief introduction to the basics of artificial neural network with its architecture. The back propagation training algorithm for a multi layer neural network has been discussed in detail, with the necessary training procedures and factors. The artificial neural network has been designed to suit its application for the river sediment discharge. For the eleven locations of the study area, artificial neural network has been developed to predict the sediment discharge for a particular time period with the necessary input data (day, month, year, water discharge, sediment discharge and rainfall) available. Chapter 4: This chapter details the procedure for the implementation of river sediment discharge using hybrid fuzzy neural networks. The general back ground along with the operators used in fuzzy systems, is presented in brief. The fundamentals of neuro-fuzzy systems and its model have been discussed. The architecture of the multi layer fuzzy neural network has been analyzed and designed in a manner to predict the sediment discharge for the eleven locations of the study area. Both crisp and fuzzy inputs are used for the training ofthe hybrid fuzzy neural network. Chapter 5: This chapter summarizes and validates the performance and analysis of the proposed artificial neural network and hybrid fuzzy neural network for the river sediment discharge of the Narmada river basin for various locations and the results are analyzed. The chapter mainly concerns with collecting all the necessary data such as, rainfall, water discharge and sediment discharge of Narmada river basin and applied to the developed networks presented in chapter 3 and 4 to determine the prediction of sediment discharge for a particular period of time. The results obtained from both artificial neural network and hybrid fuzzy neural network are compared and analyzed based on the percentage and absolute errors between targeted and predicted sediment loads for the weekly random test cases of the Narmada river basin for the eleven locations and are tabulated. Chapter 6: This chapter gives the overall conclusions and findings of the research work. The entire thesis is substantiated with 26 tables and 89 figures. In addition, references cited in the text and appendices are also included.|
|Appears in Collections:||DOCTORAL THESES (Earth Sci.)|
Files in This Item:
|NEURO-FUZZY MODEL APPLICATION TO SEDIMENT DISCHARGE IN NARMADA RIVER.pdf||127.96 MB||Adobe PDF||View/Open|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.