Análise de viabilidade econômica de projetos de geração fotovoltaica com a estimação de índices macroeconômicos por redes neurais artificiais com seleção de atributos

Abstract

Large-scale photovoltaic energy generation projects have a long payback time, which is based on macroeconomic indicators that varies over time, nevertheless, for economic feasibility analysis purposes, are considered static. Artificial Neural Networks (ANNs) have been shown to be a useful tool for financial analysis and it is possible to find in the correlate literature successful applications for the estimation of stock prices and indices based on the conjunction of a combination of previous data. Based on these ANN applications, this paper proposes the use of multi-layer perceptron ANNs with feature selection for the estimation of a macroeconomic index used in the economic feasibility analysis of photovoltaic generation projects. Correlation-Based Feature Selection was used to select the most relevant attributes for the input of the multi-layer perceptron ANN with back propagation and this ANN was used in the estimation of the Extended National Consumer Price Index (IPCA) throughout of time so that it could be used in the analysis of economic feasibility. After data collection and training of the ANN, which presented a good performance regarding the prediction model and obtained a superior performance when the feature selection was not considered. However, the obtained errors, although small, greatly influence the economic feasibility analysis, especially for short and medium term planning horizons due to the relatively recent database used for development of this model.