Developing data-driven models for energy-efficient heating design in office buildings

Abstract

Data-driven methods have been widely applied in the prediction of energy consumption in buildings. However, existing well-established data-driven models can hardly be used for energy-efficient design. This study aims to explore the underlying causes and propose an innovative method to exclusively develop models for energy-efficient design. First, a conventional modeling process was implemented, which includes data precession, statistical analysis, feature selection, and Random Forest classification. Second, an innovative two-step method was proposed to develop data-driven models for energy-efficient design. The first step involved identifying important designable features that can be designed through classification. The second step involved developing classification models for developing energy-efficient design. The experiments were performed on the Commercial Building Energy Consumption Survey (CBECS) dataset that contains 6720 non-residential buildings. The models were built with conventional methods to realize high classification accuracy. However, they cannot be used for energy-efficient design because they lack design variables such as the thickness of wall insulation. The main contributions of this study include the identification of important designable features and development of data-driven models exclusively for energy-efficient design. The proposed method can benefit designers in developing useful data-driven models for building energy-efficient design.