Enhancing Energy Efficiency in Residential Buildings at the Design Stage Through Statistical and Machine Learning Models

Abstract

The high proportion of energy consumed in buildings has led to significant environmental problems that negatively impact human existence. It is noted that the construction of energyefficient buildings can help reduce the overall energy consumed in buildings. The prediction of building energy use is largely proclaimed to be a method for energy conservation and improved decision-making towards decreasing energy usage. Statistical and Machine Learning (ML) methods are recognised as highly effective for producing desired outcomes in prediction tasks. Consequently, ML has been extensively applied in studies focusing on the energy consumption of operational buildings. However, few studies explore the suitability of ML algorithms for predicting potential building energy consumption during the early design stage to facilitate the construction of more energy-efficient buildings. This research developed a back-to-front model for building designers, using statistical and machine learning algorithms. Embracing a positivist paradigm due to its objective stance, allows for a rigorous investigation and experimental analysis of hypotheses and objective evaluation of various models. This research includes evaluating different feature selection impacts on models for classification and regression tasks, assessing various statistical and AI tools across several criteria within the building energy research domain, and comprehensively reviewing studies on various factors influencing energy use in buildings, among other investigations and analysis. A key finding is that Gradient Boosting (GB) is identified as the most effective model in terms of both accuracy and computational efficiency. Through extensive investigation and analysis, GB emerged as the optimal choice for building an energy prediction model. A significant contribution of this research is the development of a back-to-front model that allows building designers to specify target energy consumption values. By inputting values for relevant parameters into the optimization model, designers can obtain optimal values or specifications for building features required to achieve the desired energy consumption outcomes. Remarkably, the model produces results in less than five minutes. This will essentially revolutionize energy assessment at the conceptual stage of building development sustainably. This research not only advances the theoretical understanding of building energy consumption prediction but also engenders practical tools for architects, engineers, and stakeholders to develop a more sustainable and efficient building. The integration of such a model using statistical and machine learning approaches into the design stage marks a significant step towards attaining environmentally friendly and economically viable buildings.