Detecção de determinismo e modelagem preditiva de séries temporais de consumo de energia elétrica

Detalhes bibliográficos
Ano de defesa: 2008
Autor(a) principal: Tales Argolo Jesus
Orientador(a): Não Informado pela instituição
Banca de defesa: Não Informado pela instituição
Tipo de documento: Dissertação
Tipo de acesso: Acesso aberto
Idioma: por
Instituição de defesa: Universidade Federal de Minas Gerais
UFMG
Programa de Pós-Graduação: Não Informado pela instituição
Departamento: Não Informado pela instituição
País: Não Informado pela instituição
Palavras-chave em Português:
Link de acesso: http://hdl.handle.net/1843/BUOS-8CUJ8C
Resumo: Long-term load forecasting is a task of great relevance in the context of energy market planning of a country or a region. Reliable forecasts of energy consumption make possible the definition of strategies that can promote the expansion of the electrical systemthat supplies a population into an specific climatic, cultural, political and socialeconomic context. The presentworkdealswith the application of Surrogate Data Analysis in an attempt to detect determinism in the energy consumption time series of Minas Gerais State (Brazil) and New England State (United States of America), and also with the use of mathematical modelling tools typical in Systems Identification to perform long-term load forecasting of such time series, considering a maximum prediction horizon of 60 months. In the stage of data analysis it is also presented a methodology to decomposetime series called Singular Spectral Analysis, which has been shown to be efficient in the extraction of time series characteristics, such as the trend components. Three were the mathematical tools chosen to construct the predictive models: (i) multiple auto-regressive-like models; (ii) multi-layer perceptron artificial neural networks; and (iii) local linear neuro-fuzzy networks. Besides them, it was also defined a trivial predictor, based on the previous knowledge of the series trend and seasonalitiybehavior, with the objective of using it as a tool to validate the methods used to deal with the forecasting problem. As far as determinism detection is concerned, the application of the surrogate data test has revealed the existence of inter-cycle and intra-cycle dynamics in i) the timeseries of the electricity consumption of the state ofNewEngland, ii) in those time series of specific sectors of little impact in the total electricity consumption of the state of Minas Gerais and iii) in the aggregated time series of the same state when the sampling time is smaller than one month. Such results have reflected in the modelling stage by revealing that the similarity among the predictions and the real consumption valueswas more evident when working with the New England time series than with CEMIG time series. In terms of error measures, for both the New England and the low voltage CEMIG time series, the best model in each of the three categories of models used in this work resulted in mean absolute percentual errors (MAPE) smaller than 3%, when considering 60 and 24 month prediction horizon. In the case of CEMIG total consumption time series, the MAPE was smaller than 5% for a 24 month prediction horizon when ARM models are used. The only situation in which the trivial predictor was superior than the other prediction strategies was for CEMIG low voltage time series, in which the existence of intra-cycle and inter-cycle determinism via surrogate data analysis was not verified. In such a case a simple trend and seasonality fitting wassufficient to obtain very accurate prediction results (MAPE smaller than 1,99% to a 24 months prediction horizon).