Aplicação de técnicas de mineração de dados para predição do consumo de energia elétrica em ambiente doméstico
Data
Autor(es)
Orientado(es)
Título da Revista
ISSN da Revista
Título de Volume
Editor
Abstract
This monograph aims to apply Data Mining techniques, such as ANN, KNN and Random Forests, for the prediction and analysis of results, in order to make possible decisions, in the area of electrical energy in the domestic environment. The database used includes information such as: electrical energy consumption in domestic environments and meteorological data, and the residence, which is the object of study, is located in Belgium. The records were collected and annotated by Candanedo, Feldheim and Deramaix (2017). Mathematical methods were applied, such as: correlation analysis to understand the relationships between attributes/variables, and normalization to improve the visualization of the database, standardizing the data scale. The evaluation and validation of the results obtained were carried out using metrics such as MAE, MAPE and RMSE. It is noteworthy that the results of this study surpassed those obtained in the base research, presenting results of 22.83 for MAE, 22.14% for MAPE and 54.52 for RMSE. And the base survey presented results of 31.36 for MAE, 29.76% for MAPE and 70.74 for RMSE. For future work, several opportunities stand out to improve and expand the results obtained. Check if WEKA treats Overfitting. Application of new prediction techniques and new validation methods. Another possible approach: analysis of the types of household appliances used and the times of greatest energy demand, which can be analyzed at different periods and intervals, such as weeks or months, with a focus on identifying energy savings and making national comparisons, if all households saved. Finally, replicate the study in the Brazilian context.
