Métodos de previsão de consumo de energia elétrica residencial em grande volume de dados

Abstract

Electricity is one of the primary sources of energy used by humanity. Growing concern for the preservation of the environment has stimulated the search for renewable energy sources capable of reducing impacts on nature. Population growth and the increasingly frequent use of electronic devices in almost all daily activities demand the most efficient use of electricity. Due to these challenges, it is essential to carry out planning to dimen-sion the structure of generation and transmission of electric energy. One of the tools capable of assisting in this sizing is the demand forecasting. Another major challenge in this area lies in the realization of these forecasts in large data scenarios (Big Data). This work aims to evaluate the performance of two prediction methods, ARIMA andHolt-Winters, using temporal series applied to a large volume of data. The database was provided by the DEBS 2014 Grand Challenge event, which contains electricity consumption data for a large number of households for one month. For the application of the prediction methods, we used libraries in the R language. In order to process data,the Apache Spark framework was used in conjunction with the R language to parallelize the data reading processing and filtering parameters. The treated data were convertedin to time series with hourly consumption values, throughout the month comprised by theoriginal database. Predictions were made for the region of the households as a who leand each residence individually. The results showed an advantage of ARIMA versusHolt-Winters in the scenario used, using the RMSE metric as a comparative basis of performance. However, based on similar experiments found in the literature, with due proportions, both RMSE values are within an acceptable range.