Modeling coordinated operation of multiple hydropower reservoirs at a continental scale using artificial neural network: the case of Brazilian hydropower system

ABSTRACT Reservoirs considerably affect river streamflow and need to be accurately represented in environmental impact studies. Modeling reservoir outflow represents a challenge hydrological studies since operations vary with flood risk, economic demand aspects. The Brazilian Interconnected Energy System (SIN) is an example of unique complex system coordinated operation composed by more than 160 large reservoirs. We proposed evaluated integrated approach simulate daily outflows from most the SIN reservoirs (138) using Artificial Neural Network (ANN) model, distinguishing run-of-the-river storage testing cases whether level data were available as input. Also, we investigated influence input features (14) on simulated outflow, related water balance, seasonality, demand. As result, verified that outputs ANN model mainly influenced local balance variables, such inflow present day before. However, other 4 represent different regions country, which infers about hydropower through availability, seemed some extent estimates. This result indicates advantages rather looking at each individually. In terms it was tested scenarios (WITH_Qout) without (NO_Qout SIM_Qout) observed model. NO_Qout trained while SIM_Qout all features, but fed levels observations. These 3 models compared two simple benchmarks: equal before (STEADY) same (INFLOW). For reservoirs, not necessary virtually inflow. estimates reached median Nash-Sutcliffe efficiencies (NSE) 0.91, 0.77 0.68 for WITH_, NO_ respectively, NSE 0.81 0.29 STEADY INFLOW benchmarks respectively. conclusion, presented satisfactory performances: when observations are available, WITH_Qout outperforms STEADY; otherwise, outperform INFLOW.