From Climate Data to Building Form: Machine Learning–Guided Envelope Design for Energy-Efficient Architecture

Introduction
Architectural envelope design is central to building energy performance, yet in practice it is often driven by heuristic rules and fragmented simulation workflows. This paper presents a machine learning-guided approach that translates high-resolution climate data into informed decisions on envelope geometry, material layering, and façade articulation, enabling architects to navigate complex interactions between solar gains, thermal inertia, daylight, and ventilation while preserving design freedom.

Methods
Typical Meteorological Year data from five Köppen climate classes (Cfa, Cfb, BWh, Dfa, Dfb) are processed into climate features (solar radiation clusters, diurnal temperature swings, humidity profiles, wind roses). These drive parametric envelope descriptors: orientation, window-to-wall ratio (15–60%), shading depth (0–1.5m), insulation thickness (50–300mm), and thermal mass distribution. A 12,000-variant dataset is generated using EnergyPlus and Radiance simulations. Gradient boosting and neural network models, validated via 5-fold cross-validation, predict energy use intensity (R²=0.91, RMSE=8.2 kWh/m²), overheating hours (R²=0.88), and daylight autonomy (R²=0.93) from combined climate–envelope feature vectors.

Results
Tested on office and mixed-use prototypes, the ML-guided workflow identifies envelope strategies reducing annual heating and cooling demand by 20–35% compared with ASHRAE 90.1-2019 code-minimum baselines (prescriptive envelope requirements, identical internal loads and schedules). SHAP-based sensitivity analyses on held-out data reveal climate-specific design drivers—shading geometry and solar control glazing in hot/arid contexts versus airtightness and insulation continuity in cold climates—providing interpretable, actionable guidance.

Conclusions
Machine learning can act as a climate-literate intermediary between raw weather data and envelope form-making. By embedding predictive models into parametric tools, architects gain rapid, intelligible feedback that elevates the building envelope from stylistic afterthought to a primary instrument of energy-efficient architectural design.