A machine learning algorithm to improve building performance modeling during design.

Area of Science:

• Building Science
• Computational Design
• Human-Building Interaction

Background:

• Building performance models (BPMs) are crucial for optimizing building design factors like energy efficiency and occupant comfort.
• Current BPMs often lack accurate methods to incorporate human-building interactions specific to new designs, leading to performance discrepancies.
• Imprecise modeling of human behavior in buildings hinders the optimization of design for real-world performance.

Purpose of the Study:

• To propose a novel computational framework to enhance the predictive accuracy of BPMs.
• To integrate context-aware, design-specific human-building interaction data into existing BPMs.
• To enable BPMs to account for human-building interactions relevant to novel architectural designs.

Main Methods:

• Development of a computational framework utilizing artificial neural networks (ANNs).
• Integration of existing BPMs with context-aware data on human-building interactions captured via immersive virtual environments (IVEs).
• Implementation of a feature ranking technique to analyze the impact of contextual factors on human-building interactions.

Main Results:

• An augmented BPM capable of predicting building performance with consideration for design-specific human-building interactions.
• Demonstration of enhanced estimation performance for BPMs by incorporating human-building interaction data.
• Quantification of the influence of contextual factors on human-building interactions within the design process.

Conclusions:

• The proposed framework significantly improves BPM accuracy by integrating human-building interaction data.
• Designers can leverage the framework to better understand and optimize human-building interactions for improved building performance.
• This approach bridges the gap between predicted and actual building performance by accounting for user behavior.