Flood Modeling in River System Using Gamma Memory

Abstract

AbstractIn this study, applications can be seen regarding artificial neural network (ANN)-based models in predicting flow depth and flow rate for multiple bounding sections in river basin. Flow rate and depth are measured contemporaneous for Tar river basin section for the period of two months are forecasted by using a single ANN and as many ANNs as that of sections. Multiple-inputs multiple-outputs MIMO networks: MIMO-1 ANN and MIMO-2 ANN and multiple-inputs single-output MISO ANN are applied for obtaining flow forecasts for a number of sections in a river basin. This model is based on explicitly and implicitly learning storage properties during unsteady flow in river reaches. MIMO-1 and MIMO-2 ANN learn fractional and actual storage variation characteristics, respectively, during a period and obey continuity norms, while MISO ANN does not obey the continuity principle and learns an arbitrary storage variation to forecast concurrent flows in a river system. To test applicability of MIMO-1, MIMO-2 and MISO model forms, ANNs having memory and no memory are used to forecast concurrent flows rates and flow depths for multiple sections in Tar river basin, USA. Flow rates and flow depth values for four different sections in Tar river basin, USA, are forecasted using: multilayer perceptron (MLP), time delay neural networks (TDNN) and multiple gamma memory neural network (MGMNN) models. To develop the forecasting models that learn storage variation in the reach explicitly from the storage characteristic data of the data sets, viz. 'rate of storage change' and 'average flow depth' data for the river systems. Which are used along with flow rate/flow depth data in training the networks. Model performances are evaluated by using root mean squared error, coefficient of efficiency and the peak flow criteria.KeywordsRiver systemsGamma memoryArtificial neural network (ANN)Time delay neural network (TDNN)