Otimização estocástica implícita de redes neurais artificiais para auxílio na operação mensal do sistema Coremas - Mãe D'Água.

Abstract

This paper presente monthly operating rules based on Implicit Stochastic Optimization (ISO) and Artificial Neural Networks (ANN) for a water systern located in Paraíba's outback, Brazil. The ISO technique consists of optimizing the system operation using a set of possible scenarios as input and, after, utilizing the optimal outcomes in order to construct reservoir operating rules. In this study, ANN were used for relating reservoir releases to inicial storage, current inflow, monthly estimations of potencial evaporation and demand, and previous reservoir release. The synthetic scenarios of reservoir inflows were generated by the Fragment Method (FM). The results obtained by the MF indicate that this approach has potential for simulating monthly flows in semiarid regions. The monthly operatirlg tules obtained by the ISO-ANN modem were applied to the operation of Coremas - Mãe d'Água reservoir and sustainability criteria were used for analyzing the results. The outcomes suggest the ISO-ANN model is superior to the standard tules of operation and similar to the application of a determinist modem with the knowledge of inflows for the whole operating horizon. As a consequence, this model may support the decision-making process for monthly operation of reservoirs in semiarid regions.