STREAM FLOW FORECASTING USING ANN

Abstract

The ability of human brain to perform difficult operations and to recognize complex patterns has fascinated scientists for centuries. The term neural network describes a broad category of algorithms that mimic the way human brain seems to receive, process, store and communicate data. The areas addressed by neural networks include data compression, optimization, pattern matching, system modeling, and function approximation. Methods of flow forecasting are generally model based. Physical processes, which influence the occurrence of flow in streams, are highly complex and uncertain, therefore it becomes even more difficult to fit them in some form of deterministic or statistical model. Such problems, however, can be tackled efficiently by the Artificial Neural Network (ANN) approach because of its in-built mechanism of growing wiser with experience. ANNs are capable of adapting the complexity of input-output patterns and accuracy goes on increasing as more and more of data are available to them. This study discusses the development and application of an artificial neural network based model for stream flow forecasting in the Negara river basin, Indonesia. A three-layer network is developed having one input, one hidden and one output layer. Training is conducted using feed-forward back propagation algorithm where the inputs and target outputs are presented to the network as a series of learning sets. Inputs given to the network are mean daily rain fall and average daily flow, where as the target output is the next day flow. The rain fall and flow data of years 1978 to 1983 is used for training the network and 1984 data is used for testing. After training is completed the neural network is used to forecast flow with a lead time of one day using test inputs of current rain fall and flow. Results show that the performance of the network to predict flow of one day ahead varies from 0.679 to 0.994 in terms of efficiency coefficient, and thus the ANN could be used as an alternative to the conventional stream flow forecasting methods.