Extreme Learning Machine with Evolutionary Parameter Tuning Applied to Forecast the Daily Natural Flow at Cahora Bassa Dam, Mozambique.

This paper proposes a hybrid approach combining an Extreme Learning Machine and a Genetic Algorithm to predict the short-term streamflow at the Cahora Bassa dam, the largest hydroelectric power plant in southern Africa. To predict the streamflows seven days ahead, the model uses as input the past river flows, information from humidity, rainfall, and evaporation measures from the lake upstream of the dam. The choice of the Extreme Learning Machine’s internal parameters, crucial for excellent model performance, is performed by a Genetic Algorithm. A set of five metrics was used to assess the performance of the hybrid approach. The computational experiments show the proposed approach outperforms other machine learning methods such as ElasticNet linear model, Support Vector Machines, and Gradient Boosting. However, the ELM prediction model overestimates higher flows. The approach arises as a practical tool to predict the streams which have the potential to help the dam operations balancing the needs of energy production and the safety of the population living downstream of the dam.