Metodologia DWEXI para previsão de energia eólica, utilizando modelos híbridos e ensemble para a região sul do Brasil

Abstract

Wind power generation forecasting is a strategic element for electrical system planning, especially in Brazil, where climate variability may lead to increased operational costs and greater dependence on more expensive or polluting energy sources. In this context, this dissertation aims to develop and evaluate hybrid approaches for forecasting wind power generation in the Southern region of Brazil, seeking greater predictive accuracy. The research was structured into three main stages. The first consisted of a systematic literature review conducted using the Methodi Ordinatio 2.0 method, with the objective of identifying methodological trends, preprocessing techniques, and the most commonly used performance metrics. The results showed the absence of a universally superior method, as well as the widespread use of hybrid models and the MAE and RMSE metrics for performance evaluation. In the second stage, individual, hybrid, and ensemble models were evaluated using the monthly time series of wind power generation in the Southern region of Brazil, considering a 12-month forecasting horizon. The results demonstrated the superiority of hybrid models over individual models, with the Additive Holt-Winters with Multiplicative model standing out as presenting the best predictive performance. The third stage consisted of the development of an innovative hybrid approach, named DWEXI, applied to hourly wind power forecasting in a wind complex located in Chuí. The method integrates data preprocessing techniques, neural network- and decision tree-based models, as well as an intelligent weighting mechanism for forecast combination. The results indicated that DWEXI achieved superior performance in most of the analyzed forecasting horizons, especially in the medium and long term. It is concluded that the proposed approach is effective for wind power generation forecasting, constituting a relevant decision-support tool for the energy sector and contributing to electric matrix planning and operational risk mitigation.