Aplicação de técnicas de inteligência computacional para a previsão de cargas de aquecimento e resfriamento em edificações

Abstract

Due to climate change, by 2040 buildings may consume 20% more energy, with energy performance being the key element for achieving sustainable development in the building sector. Design costs, environmental damage and constructive characteristics, such as geometry, thermal properties of materials and weather conditions, are necessary to assess the energy performance of a building. One way to solve this problem is to apply Machine Learning Methods to estimate a response from data. The use of Machine Learning Methods can assist construction specialists in scenario analysis at an early stage of the project and bring savings to the building. The present work evaluates the application of the Elastic Net, Extreme Learning Machine and Extreme Gradient Boosting models for the forecasting of heating and cooling loads in residential buildings. Two databases are used to evaluate the performance of the methods, including geometric input variables and two thermal output variables. For the selection of method parameters, the Differential Evolution optimization algorithm was applied to find the sets of hyperparameters that reinforce the predictive capabilities of the models. Comparisons of the results were made using the MAE, MAPE, RMSE and R2 metrics. The results showed that the Extreme Gradient Boosting achieved a better performance among the tested methods and also compared to the literature, presenting the lowest values for the error metrics and significant difference in the statistical tests. In addition, the Differential Evolution algorithm showed to be effective for the optimization of the parameters of the tested models, and may also be applied to other models in the literature. Thus, a combination of Differential Evolution and Extreme Gradient Boosting methods can be applied to predict thermal loads in buildings, assisting in projects that aim at energy saving and sustainability.