Forecasting of Water Flow in a Hydroelectric Power Plant Using LSTM Recurrent Neural Network

The water flow is an important information for the study of the energy production as well as management and hydropower control. The forecast of future information scenarios allows to take advanced actions in order to optimize electricity generation. This work proposes a model to generate a forecast of water flow based on a Recurrent Neural Network, more specifically the long short-term memory (LSTM) type. The dataset used to validate the LSTM model is obtained from flow history of Jirau Hydroelectric Power Plant, installed on the Madeira River in the state of Rondônia, Brazil. The model was trained and tested on one-day time-step resolution. Experimental results shows a lower error according its predictive capacity. The resulted forecast was evaluated by operating staff ex-perts, from Jirau Hydroelectric Power Plant, which attested the results can be used in real operation scenario. It is concluded that LSTM model is a good strategy for the forecast of water flow for the study of hydroelectric turbine efficiency.