Modelo para previsão de demanda de eletricidade com redes neurais artificias integradas a métodos multicritério
In order to avoid excessive electricity production or below demand, efficient planning and adoption of policies are essential. For this, it is necessary to improve the techniques for forecasting electricity demand in a country, as a production that does not meet the demand for cause. On the other ha...
| Autor principal: | Deina, Carolina |
|---|---|
| Formato: | Dissertação |
| Idioma: | Português |
| Publicado em: |
Universidade Tecnológica Federal do Paraná
2020
|
| Assuntos: | |
| Acesso em linha: |
http://repositorio.utfpr.edu.br/jspui/handle/1/4921 |
| Tags: |
Adicionar Tag
Sem tags, seja o primeiro a adicionar uma tag!
|
| Resumo: |
In order to avoid excessive electricity production or below demand, efficient planning and adoption of policies are essential. For this, it is necessary to improve the techniques for forecasting electricity demand in a country, as a production that does not meet the demand for cause. On the other hand, excessive production generates waste and high costs. This work aimed to improve the techniques of the electricity demand forecasting process, using a model that integrates neural networks and the ELECTRE I multicriteria method. In the literature, see if you want to include independent variables in the forecasting model, the results improve significantly, but there is still a lack of efficient models that help in this process. Thus, the model presented has five distinct stages: 1) selection of input variables by the ELECTRE I method, whose objective was selected only as causal variables, excluding as variables that can only be correlated, but without direct influence demand behavior; 2) pre-processing of data with identification of demand behavior, considering the possibility of seasonal display, trend, cyclicality and random terms and non-stationarity; 3) demand forecast executed based on the linear models of SEHW, AR and ARIMA and three models of Artificial Neural Networks, MLP, RBF and ELM; 4) data post-processing with data transformation in the original dimensions; 5) Comparison of models used based on MSE, MAE and MAPE error measures. To test the proposed methodology model, select a historical series of electric energy consumption in Paraná. How they were performed considering 1, 3 and 6 steps ahead. In addition, the inputs used by the neural models were selected using the Wrapper method. As a result, or ELECTRE, I select as explanatory variables of temperature and average evaporation, when these were included in the constant predictive models, as improvements in the results of all as RNAs. In addition, for this problem, how RNAs performed better than linear models. It stands out as ELM and RBF Networks as the best editors. |
|---|