Análise de séries temporais com uso de redes neurais artificiais em dados meteorológicos para previsão de chuva e eventos climáticos severos

• Autor(a) principal: Soares, Augusto Carvalho
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFU
• Texto Completo: https://repositorio.ufu.br/handle/123456789/36857; https://doi.org/10.14393/ufu.di.2022.680
• topic: Séries Temporais; Redes Neurais Artificiais; Previsão; Chuva; Time Series; Artificial Neural Networks; LSTM; Weather; Forecasting; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO; Computação; Redes neurais (Computação); Instrumentos meteorológicos; Chuvas - Frequência da intensidade

Weather forecasting and climate prediction have become some of the most useful resources in the modern world. Global warming is a serious problem. The Climatological Normals in Brazil are changing in a worrying way. A recent INMET report (2022) brings studies of changes in the last 20 years, showing an increase of 1.6° Celsius in the average temperature. Long-term projections point to a scenario in which ecosystems and production chains may collapse. These changes can already be seen in the increasing occurrence of severe weather events. It is essential to know in advance when such events will occur in order to better face their consequences. The mitigation of natural disasters can be achieved through the anticipation of climatic conditions and the previous warning to the population by the authorities before an imminent severe event. Science searches for new and accurate methods for correct weather forecasting, some proposed solutions indicate the use of ANNs for classification, however, there is still little known about regression tasks for non-traditional forecasting methods. This work proposes the application of regressive ANNs for the task of rainfall volume forecasting. The models were submitted to a battery of tests in “brute force” style, to find their best parameters. The results obtained were 80% and 81% of correct for LSTM and MLP ANNs respectively, proving the hypothesis that weather forecasting is possible using AI and ANNs.