Egresso do PPGEM dará palestra na Universidade da Califórnia em Los Angeles (UCLA) sobre energia eólica
O PPGEM convida a comunidade acadêmica para a palestra do aluno Djayr Bispo, atualmente realizando doutorado-sanduíche na University of California em Los Angeles (UCLA) sobre previsão de energia eólica utilizando modelos de aprendizado profundo. Seguem abaixo os dados da palestra:
Título: Short-Term Wind Power Forecasting Using Transformer-Based Model Architecture”
Resumo: The efficiency of the global energy grid must be maximized in order to meet the increasing demand for electrical energy and lessen its effects on the environment. In the wind industry, precise power generation projections are essential for boosting wind farm productivity and optimizing maintenance and operation plans. The short-term power forecasting of wind turbines operating in Northeast Brazil is the main topic of this study. Deep learning networks and other reference models were tested, with a focus on those using the Transformer architecture. The proposed framework includes the adaptation of the original Transformer architecture to time series forecasting by means of two significant changes are introduced. First, a richer representation of temporal data is made possible by the encoder’s use of the Time2Vec layer. The second is the replacement of the traditional FullAttention mechanism with the FlowAttention and FlashAttention mechanisms. The results show that the changes performed better than the benchmarking models, making them more appropriate for forecasting wind turbine energy production. Therefore, this work contributes towards the goal of maximizing electrical energy generation and optimize the productive efficiency of wind farms.
Informações do Evento
• Data: 21/11/2024
• Horário: 1-2 pm, PT
• Local: https://ucla.zoom.us/j/92451782922