Wind Power Forecast: Ensemble Model Based in Statistical and Machine Learning Models

• Autor(a) principal: Gebin, Luís Gustavo Gutierrez
• Data de Publicação: 2020
• Outros Autores: Salgado, Ricardo Menezes, Nogueira, Denismar Alves
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Research, Society and Development
• Texto Completo: https://rsdjournal.org/index.php/rsd/article/view/11251
• topic: Energia eólica; Modelos estatísticos; Aprendizado de máquina; Seleção de variáveis.; Energía eólica; Modelos estadísticos; Aprendizaje automático; Selección de variables.; Wind energy; Statistical models; Machine learning; Selection of variables.

The energy sector is one of the pillars of modern society, considering the fact that it can impact the development of a country in several segments – economic, social, environmental etc. As for the environmental impact, it is known that its production can generate waste and cause harm to the environment, causing an increase in the greenhouse effect and contributing to global warming. In this sense, the wind energy sector stands out, based on a production that is derived from the wind and is considered “clean”. However, there is much uncertainty regarding its production, as it is not possible to produce or store wind. Because of this, prediction models are made necessary, so that there is security in the utilization of this type of production. Predictions are not trivial tasks, since there are several variables that impact their achievement, as well as their stability or (instability). Several models have already been used for this purpose, such as the traditional ones (ARIMA), the intelligent ones (XGBoost and Random Forest) and the ensemble models (joint/combination models), which have been gaining prominence. The goal of the work is to develop a strategy to forecast the production of wind energy on an hourly basis, in two different stations and with different models. After carrying out the experiments, it can be seen that the individual models (especially XGBoost) present good results; however, it should be noted that, in most outlier cases, these models suffer from a problem of overestimation. The ensemble model, however, managed to gap this deficiency in relation to the individual models, correcting overestimation cases.