Application of Classification and Feature Selection in Building Energy Simulations

Abstract: Building energy performance is one of the key features in performance-based building design decision making. Building envelope materials can play a key role in improving building energy performance. The thermal properties of building materials determine the level of heat transfer through building envelope, thus the annual thermal energy performance of the building. This research applies the Linear Discriminant Analysis (LDA) method to study the effects of materials' thermal properties on building thermal loads. Two approaches are adopted for feature selection including the Principal Component Analysis (PCA) and the Exhaustive Feature Selection (EFS). A hypothetical design scenario is developed with six material alternatives for an office building in Los Angeles, California. The best design alternative is selected based on the LDA results and the key input parameters are determined based on the PCA and EFS methods. The PCA results confirm that among all thermal properties of the materials, the four parameters including thermal conductivity, density, specific heat capacity, and thickness are the most critical features, in terms of building thermal behavior and thermal energy consumption. This result matches quite well with the assumptions of most of the building energy simulation tools.