Análisis de eficiencia energética en edificios residenciales utilizando aprendizaje de máquinas basado en regresión lineal

Abstract

The objective of this curricular integration work was to apply automatic learning analysis techniques (ML). It was based on regression methods to predict heating and cooling loads in residential buildings, through the generation of regression equations and Python language code for ML techniques. Therefore, for the creation of the learning algorithm, an energy performance data set was used whose database was extracted from the Machine Learning Repository of California University, Irvine. This data set contains eight attributes (or characteristics, denoted by X1 through X8) and two responses (or outcomes, denoted Y1 and Y2) whose purpose is to use the eight functions to predict each of the two responses. The first part of the work is dedicated to the exploratory analysis of data and visualizations, as well as the application of statistical techniques and analysis of the main components of the data set. After this analysis in a second part, several regression methods were developed, where the performance of each algorithm was compared, with the aim of finding the most optimal model to predict energy efficiency in residential buildings. Finally the different regression algorithms, analyzing the results, the precision and the errors that these showed. The Random Forest Regression, Decision Tree Regressor, Extra Tree Regressor and the linear regression model were determined that are models applicable to the prediction of the value of the heating and cooling load. Because, the four models show accuracies greater than 90%, the Random forest model being the most suitable for this study as it gives an accuracy of 98%. It is recommended to carry out other ML models with the purpose to improve the prediction results.