Assessing the impact of hydropower projects in Brazil through data envelopment analysis and machine learning

Abstract

The aim of this study was to assess the environmental impact of hydroelectric power generation projects and classify them according to their scale of environmental impact. To achieve this objective, the combination of Data Envelopment Analysis (DEA) and Artificial Neural Networks (ANN) techniques was applied to 53 hydroelectric power plant projects in the evaluation phase in Brazil. The main results were: a) the proposed index indicates that 7 of the 10 worst hydroelectric projects are of the Large Hydropower Plant (LHP) type; b) the neural model for predicting the environmental impact of hydroelectric projects has an error of less than 0.001; c) the neural model for classifying hydroelectric projects in terms of their environmental impact reached a performance of 99.0% accuracy. In general, this study contributes to the use of a hybrid decision-making approach based on a combination of DEA-ANN for energy policies, in addition to enabling an improvement in the evaluation of hydroelectric generation projects.