Image-based Automatic Modeling and Performance Evaluation for House Plans in the Early Design Stage

Abstract: The performance evaluation is crucial in comparing multiple house plans during the early stages. However, simulation-based methods often face challenges such as time consumption and complex modeling processes. Therefore, this paper extracted the features from house plans and proposed an image-based algorithm for automatic geometric simulation modeling. The algorithm successfully modeled 16,000 house plans and obtained simulation results for their energy consumption, indoor thermal comfort, and daylighting. This paper further compares the MARS-based and XGBoost-based surrogate models for performance evaluation using the house plan dataset. The results indicate that XGBoost performed better on all metrics.