A new approach to river flow forecasting: LSTM and GRU multivariate models

Abstract

Hydroelectric power stations are responsible for renewable energy generation, especially in countries with many rivers as in Brazil. It is very important to have good estimates of the hydrological flow in order to determine if thermoelectric power plants should begin operation, an event that increases the costs of electricity and also has a terrible environmental impact. The monthly flow of a river was estimated using two recurrent neural networks techniques: Long-Short Term Memory (LSTM) and Gated Recurrent Unit (GRU). The results were compared with other articles that had the same structure using the same data: the Rio Grande river in the Furnas and Camargos dam.