Water Flow Forecasting Based on River Tributaries Using Long Short-Term Memory Ensemble Model

Water flow forecasts are an essential information for energy production, management and hydropower control. Advanced actions to optimize electricity production can be taken based on predicted information. This work proposes ensemble strategy using recurrent neural networks generate a forecast of water at Jirau Hydroelectric Power Plant (HPP), installed the Madeira River in Brazil. The consists combining three long short-term memory (LSTM) that model two its tributaries: Mamoré Abunã rivers. historical data from streamflow river tributaries used validate LSTM model, where each time series modeled separated by models result as input another order main river. experimental results present low errors training test sets individual model. In addition, these were compared with operational performed HPP. proposed showed better accuracy four five scenarios tested, which indicates promising approach explored forecasting tributaries.