Author(s): Farias Camilo Allyson Simoes
Linked Author(s):
Keywords: Recurrent neural network; Implicit stochastic optimization; Reservoir operation
Abstract: This paper aims at investigating the use of a recurrent neural network (RNN) forecaster for assisting reservoir operations performed by Implicit Stochastic Optimization (ISO) and Artificial Neural Networks (ANNs). The RNN architectures have recurrent connections that implicitly allow the network to detect and produce time-varying patterns, which makes them very suitable for the prediction of water resources time series. The fundamental principle of the ISO procedure is to generate synthetic inflow scenarios which are used by a deterministic optimization model to find optimal reservoir releases. The ensemble of optimal release data is then examined in order to develop operating rules for each month of the year. The ANNbased ISO procedure calculates, by means of ANNs, the release at each month conditioned on the month’s initial storage and the forecasted inflow for the month. Such monthly inflow is determined by the RNN, as a function of the previous inflow and current rainfall. The methodology is applied to the Ishitegawa Dam, which is the reservoir that supplies the city of Matsuyama in Japan. Scarcity of water is a periodical problem in this city and thus it is very important to improve the water resources in the region. ANN-ISO-based releases assuming the forecasted inflow as perfect forecast are also carried out for comparison. The excellent correlation obtained by the RNN forecaster indicates that this tool is very efficient for onemonth-ahead forecasting of reservoir inflows. The releases carried out by the ANN-ISOgenerated policies combined with RNN-based forecast were shown to be very similar to those produced by ANN-ISO-generated rules using perfect forecast and better than the ones obtained by standard rules of simulation.
Year: 2007