Modelagem do comportamento temporal do quasar 3C273 utilizando técnicas de aprendizado de máquina

• Ano de defesa: 2019
• Autor(a) principal: Ferrari, Ricardo Bulcão Valente
• Orientador(a): Botti, Luiz Claudio Lima
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Presbiteriana Mackenzie
• Palavras-chave em Português: AGN; quasar 3c273; rádio; variabilidade; curva de luz; computação evolutiva; redes neurais; Long Short Term Memory (LSTM)
• Área do conhecimento CNPq: CNPQ::CIENCIAS EXATAS E DA TERRA::FISICA; CNPQ::CIENCIAS EXATAS E DA TERRA::ASTRONOMIA::INSTRUMENTACAO ASTRONOMICA::RADIOASTRONOMIA
• Link de acesso: http://dspace.mackenzie.br/handle/10899/25801

Using machine-learning techniques, this work aims to model the temporal behavior from an extragalactic radio source, the quasar 3C273, register in well-sampled light curves, applying this result to less sampled curves, so it would be possible to fill gaps that have occurred in moments in which the observations were not possible. It is also intended to study this object from the point of view of its spectral and temporal variability in the radio range of the electromagnetic spectrum. The 3C273 quasar has been monitored in radio since 1963 (Schmidt, 1963) at various frequencies and was chosen for this work because of the large amount of data available. Data from several Radio Observatories were used, such as the Radio Observatory of Itapetinga (ROI) future ROPK – Radio Observatory Pierre Kaufmann, at 22 and 43 GHz; of the Michigan Radio Observatory (UMRAO) at 4,8, 8, 14,5 GHz and the Metsähovi Radio Observatory at 22 and 37 GHz. Millimetric and submillimetric data were also used, at 1.1 and 0.8 mm from Submillimetric Interferometer Array (SMA) and at X-rays, in the range of 2-10 keV (Trueler et al. 1999; Soldi et al. 2008). Firstly we developed a code based on genetic algorithm (GA), which is a good tool to adjust the data to the observed data. However, for the GA work properly, a previous model for the fitting function is necessary, and for this reason a study of a more appropriate technique, that do not require a previous model, was performed. The artificial neural networks (ANN) mimic the learning process of the human brain, forming and applying the knowledge gained from past experiences to new problems or situations. With the mathematical development of the neural network, new and more complex architectures could be developed. In this work, a Recurrent Neural Network type called Long and Short Time Memory (LSTM) was used, which proved to be an excellent choice to model the light curves of the 3C273 quasar and to make predictions on daily light curves.