Studying Buildings Outlines to Assess and Predict Energy Performance in Buildings: A Probabilistic Approach

TL;DR

This study investigates how building outline shapes influence energy consumption, using probabilistic models and synthetic data, revealing that square shapes are generally more energy-efficient and that polynomial regression can accurately predict energy use.

Contribution

The paper introduces a probabilistic approach to model the impact of building outline shapes on energy consumption, demonstrating high-accuracy predictions with polynomial regression models.

Findings

Square-shaped buildings are more energy-efficient than T, U, and L shapes.

T, U, and L shapes exhibit similar energy consumption behavior.

Degree 2 polynomial models predict energy consumption with over 98% R squared.

Abstract

Building performance is commonly calculated during the last phases of design, where most design specifications get fixed and are unlikely to be majorly modified based on design programs. Predictive models could play a significant role in informing architects and designers of the impact of their design decisions on energy consumption in buildings during early design stages. A building outline is a significant predictor of the final energy consumption and is conceptually determined by architects in the early design phases. This paper evaluates the impact of a building's outline on energy consumption using synthetic data to achieve appropriate predictive models in estimating a building's energy consumption. Four office outlines are selected in this study, including square, T, U, and L shapes. Besides the shape parameter, other building features commonly used in literature (i.e., Window to Wall Ratio (WWR), external wall material properties, glazing U value, windows' shading depth, and building orientation) are utilized in generating data distribution with a probabilistic approach. The results show that buildings with square shapes, in general, are more energy-efficient compared to buildings with T, U, and L shapes of the same volume. Also, T, U, and L shape samples show very similar behavior in terms of energy consumption. Principal Component Analysis (PCA) is applied to assess the variables' correlations on data distribution; the results show that wall material specifications explain about 40% of data variation. Finally, we applied polynomial regression models with different degrees of complexity to predict the synthesized building models' energy consumptions based on their outlines. The results show that degree 2 polynomial models, fitting the data over 98% R squared (coefficient of determination), could be used to predict new samples with high accuracy.