Submitted by MONARA PEREIRA DA ROSA CARVALHO null (momoprc@gmail.com) on 2017-03-17T12:47:54Z No. of bitstreams: 1 MONARA_Dissertacao.pdf: 2926386 bytes, checksum: 52ab3ee5e454a3b74043a0bbef9630de (MD5) / Approved for entry into archive by Juliano Benedito Ferreira (julianoferreira@reitoria.unesp.br) on 2017-03-21T19:15:17Z (GMT) No. of bitstreams: 1 carvalho_mpr_me_ilha.pdf: 2926386 bytes, checksum: 52ab3ee5e454a3b74043a0bbef9630de (MD5) / Made available in DSpace on 2017-03-21T19:15:17Z (GMT). No. of bitstreams: 1 carvalho_mpr_me_ilha.pdf: 2926386 bytes, checksum: 52ab3ee5e454a3b74043a0bbef9630de (MD5) Previous issue date: 2017-01-20 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / This work analysis load data from desegregated levels that presented difficulties to load forecasting with several methods due to variation in electrical energy consumption. The application proposed in this work is short-term load forecasting to a university building by GRNN (General Regression Neural Network) considering the bottom up approach and using a moving average filter to deal with the missing or wrong data. It is presented the system that provides the data as well as the methods used for pre-processing and realize the forecasting. The results are evaluated by MAPE (Mean Absolute Perceptual Error) and are considered good when compared with other methods. / Este trabalho destaca a análise de dados provenientes de locais com níveis de consumo mais desagregados que apresentam dificuldades para previsões de demanda com vários métodos devido à alta variação no consumo de energia elétrica. Apresenta-se resultados de previsões de demanda de curto prazo da energia elétrica consumida em um bloco de uma universidade por meio da rede neural de regressão generalizada (GRNN), utilizando a abordagem de modelagem de dados de baixo para cima e tratamento de ruídos e dados faltantes no banco de dados através da aplicação de um filtro de médias móveis. É apresentado o local que fornece as informações para os estudos e a etapa de pré-processamentos dos dados. Foi possível analisar a assertividade das previsões de acordo com o cálculo do MAPE, mostrando vantagens ao se comparar a outros métodos utilizados para os mesmos fins.

Orientador: Anna Diva Plasencia Lotufo / Resumo: This work analysis load data from desegregated levels that presented difficulties to load forecasting with several methods due to variation in electrical energy consumption. The application proposed in this work is short-term load forecasting to a university building by GRNN (General Regression Neural Network) considering the bottom up approach and using a moving average filter to deal with the missing or wrong data. It is presented the system that provides the data as well as the methods used for pre-processing and realize the forecasting. The results are evaluated by MAPE (Mean Absolute Perceptual Error) and are considered good when compared with other methods. / Mestre