Forecasting of the Reservoir in Flow with Recurrent Neural Network Architecture

One of the most difficult challenges has been forecasting a hydrologic time series due to the vast variety of data, the uncertainty in the parameters impacting the time series, and insufficient data. The feed-forward network and the recurrent neural network are the two selected networks. The ordered partial derivatives method trains the recurrent neural network, while the feed-forward network is trained using the conventional back propagation algorithm with several modifications. In this research, the daily flow of the stream of the water reservoir hydroelectric dam, and another natural water reservoir, is studied using deep Recurrent Neural Network (RNN) architectures. When modeling inflow in the most realistic way possible, it may boost energy production and income by optimizing water management, considering that the water reservoir is utilized for hydroelectric operations and energy generation. Hence, many percent gains can be beneficial. As a result of the precision of the deep learning techniques, RNNs can be used to estimate streamflow to a water reservoir.