Modelo híbrido de previsão de séries temporais para possíveis aplicações no setor de geração eólica

Abstract

In this paper an innovative hybrid model of time series prediction based on the combination of two functions (linear and nonlinear) of the Holt-Winters and Artificial Neural Networks models is presented. This model is applied in wind speed in northeastern Brazil, and was able to perform short and long term forecasts with good accuracy. We highlight the efficiency of the proposed model in providing perfect adjustments to the data observed, being this affirmative according to the low values found in the statistical analysis of errors, for example, with percentage error of approximately 5.0%, and also with the value of the Nash-Sutcliffe coefficient of efficiency of approximately 0.96. These results were important for the accuracy of the data, so that they could follow the profile of the observed time series, mainly revealing greater similarities of maximum and minimum values between both series, thus showing the capacity of the model to represent characteristics of local seasonality. Wind speed prediction methods can be a useful technique in the wind power sector, for example, being able to acquire important information on how local wind potential can be harnessed for possible electric power generation.