PREDICTING PRIMARY ENERGY SAVINGS OF BUILDING RETROFIT MEASURES WITH DECISION-TREE-BASED ENSEMBLE METHODS

Abstract

Primary energy is the quantity often used to express the total amount of consumed or saved energy. It is especially important for certification and standards compliance evaluation of buildings. In order to accurately assess primary energy consumption and savings of building retrofit measures, one needs adequate models that can be based on the knowledge of phenomena, on the collected data, or both. This paper analyzes the learning performance and accuracy of data-driven models based on decision trees that predict primary energy savings related to building retrofit measures. It uses supervised machine learning methods — regression decision trees and ensemble methods based on them — to train, validate, and test such models. Ensemble methods based on decision trees are powerful, accurate, but also convenient to prepare and fast to train. In addition, they calculate the relative importance of each feature. The research results with highly accurate data-driven models and consistent feature importance values.