Hugo Miguel Nogueira Mende
A Dissertation presented in partial fulfilment of the requirements of
the Degree of Master in Integrated Business Intelligence Systems
September, 2020
------------------------------------
Tese em MSIAD, classificada com 19 valores
------------------------------------
Abstract
With the development of underdeveloped countries and the digitization of societies, energy consumption is expected to continue to show high growth in the coming decades. While there is still a strong focus on fossil fuels for energy generation, the implementation of energy policies is crucial to gradually shift to renewable sources and the consequent reduction in CO2 emissions. Buildings are currently the sector that consumes the most energy. To contribute for a better energy consumption efficiency, it was proposed a framework, to be applied to buildings or households, to allow users to know their energy consumption and the possibility to forecast it. Different data analysis techniques for time series were used to provide information to the user about their energy consumption as well as to validate important data characteristics, namely stationarity and the existence of seasonality, which can have an impact in the forecasting models. For the definition of the forecasting models, state of the art was done to identify used models for energy consumption forecasting, and three models were tested for both types of data, univariate and multivariate. For the univariate data, the tested models were SARIMA, Holt-Winters and LSTM as for the multivariate data, SARIMA with exogenous variables, Support Vector Regression and LSTM. After the first execution of each model, hyperparameter tuning was done to conclude on the improvement of the results and the robustness of the models for later application to the framework.
Keywords: energy consumption, forecasting, framework, data analysis
Ler mais aqui.
Candidatar a MSIAD aqui.
