The Prediction Of Water Consumed Pattern Using Time Series Data (Recurrent Nueral Network) In Water Test Bed Grid (Network).

Abstract

There are several factors that make forecasting a hydrologic time series a difficult endeavour, including a wide range of data, the lack of accurate data, and a lack of enough data. It has recently become common practise to use artificial neural networks (ANNs) for time series forecasting in numerous industries. Forecasting river flows using artificial neural networks is demonstrated in this research. A feed forward network and a recurrent neural network have been selected for the experiment. The recurrent neural network is trained utilising the method of ordered partial derivatives, while the feed forward neural network is taught with the usual back propagation approach. Both networks' architectures and training methods are described in detail. ANN models were used to train and estimate monthly flows of an Indian river with a catchment area of 5189 km2 up to the gauging station using the models that were chosen for this task Both single-step and multiple-step forecasts may be made using the trained networks. A comparison of the two networks reveals that feed forward networks were outperformed by recurrent neural networks. In addition, recurrent neural networks had smaller architectures and took less time to train. For both single-step and multiple-step forecasting, the recurrent neural network performed better. The use of recurrent neural networks in river flow forecasting is therefore strongly advocated.