Rapid Machine Learning-Based Energy Prediction for BIPV-Integrated Modular Buildings
The accelerating global transition toward carbon neutrality demands rapid and reliable energy prediction tools for innovative building systems such as Building-Integrated Photovoltaic (BIPV) modular buildings. Conventional physics-based simulation methods, while accurate, are computationally intensive and unsuitable for real-time design optimization. This study proposes a novel machine learning-based rapid energy prediction methodology tailored specifically to the thermal and geometric characteristics of modular BIPV-integrated buildings.
Feature Engineering for Modular Building Representation
A comprehensive feature engineering framework was developed to capture the distinctive attributes of modular construction. The approach incorporates six-surface thermal property encoding, geometric parameters, and detailed solar irradiance calculations to represent envelope exposure and inter-module interactions. This structured encoding ensures that key thermal behaviors and photovoltaic influences are effectively translated into machine-learning-compatible inputs.
Modular Decomposition Strategy and System Integration
To maintain computational efficiency while preserving system-level fidelity, a modular building decomposition strategy was implemented. Individual modules are analyzed independently, and their interactions are aggregated to achieve accurate whole-building performance predictions. This hierarchical modeling approach enhances scalability and supports flexible application across different modular configurations.
XGBoost Model Performance Across Climate Zones
An XGBoost-based prediction model was trained and evaluated across four representative climate zones. The model demonstrates strong predictive capability, achieving R² values exceeding 0.93 for heating loads, cooling loads, and total energy consumption. These results confirm the robustness and generalizability of the proposed approach under diverse climatic conditions.
Experimental Validation and Computational Efficiency
Validation using a real-world BIPV-integrated modular building confirms prediction accuracy within industry-acceptable limits, with mean absolute errors below 1.5 °C for indoor temperature estimation. Computational efficiency analysis reveals prediction speeds more than 2,000 times faster than traditional simulation tools, enabling near real-time performance assessment and iterative design exploration.
Integration with Parametric Design and Practical Implications
The methodology was successfully integrated with Grasshopper parametric design environments, providing immediate energy performance feedback during conceptual design phases. By eliminating computational bottlenecks associated with traditional simulation workflows, the proposed framework supports rapid performance-driven decision-making and facilitates broader adoption of sustainable modular construction practices in energy-informed architectural design.
Comments
Post a Comment