Panorama de modelos de previsão de explosões solares utilizando aprendizado de máquina

• Autor(a) principal: Ferreira, Juliana Sabino, 1992-
• Data de Publicação: 2022
• Outros Autores: Gradvohl, André Leon Sampaio, 1973-, Silva, Ana Estela Antunes da, 1965-, Coelho, Guilherme Palermo, 1980-
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Repositório da Produção Científica e Intelectual da Unicamp
• Texto Completo: https://hdl.handle.net/20.500.12733/8684
• topic: Erupções solares; Aprendizado de máquina; Solar flares; Machine learning; Solar activity - Forecasting; Magnetic parameters; Artigo original; Atividade solar - Previsão

Abstract: Solar activities, including solar flares and coronal mass ejections, influence Space Weather and consequently affect Earth. In particular, technological systems orbiting the Earth and other systems on the ground are affected by solar radiation. Therefore, predicting solar flares helps in taking actions that aim to minimize the consequences of these phenomena in these technologies. The solar activity data is captured by specialized instruments and made available for prediction models to perform solar flare forecasting. However, the mechanism of solar flares is not fully understood. There are several models for predicting solar flares, many using machine learning. Despite being different models, we noted several common characteristics between them, which point to important factors that indicate characteristics of solar flares. One of the examples is the attributes extracted from solar data, which we classify as magnetic or morphological and help in prediction models. Thus, the research reported here sought to outline the works in the scientific literature that predict solar flares using machine learning. In this analysis, we consider some aspects, such as the algorithms and data used, as well as the types of attributes – magnetic or morphological – used. In addition, this research aims to verify the frequency of some characteristics present in these models, such as data sources, attributes, methods used, and forecast windows. The results pointed to the greater efficiency of models that use magnetic attributes to forecast solar flares. Other factors that also influence these models were the forecast windows, database and algorithms used