ML for Sustainable Architecture: A Research Portfolio

⚙️ Integrated ML & Generative Design for Climate-Adapted Energy-Efficient Buildings

• A research-driven project integrating reinforcement learning, explainable AI, generative design, and energy performance simulation to support early-stage architectural decision-making for climate-adapted, Passivhaus, and net-zero buildings.--

🚧 Project Status

This project is under active development as part of a personal research portfolio, showcasing my continuous learning and practical application of Machine Learning for a potential PhD application at institutions like Edinburgh University..

🎯 Goals

• To automate early-stage building layout generation using generative design and reinforcement learning optimization
• To integrate site conditions (EPW weather, solar orientation, etc.) and climate adaptation data.
• To optimise designs based on energy use intensity (EUI), thermal comfort, and resilience metrics.
• To evelop explainable AI models to provide transparent, interpretable energy performance predictions.
• To provide a decision-support tool for architects and designers with feedback-driven iterative design.

🛠️ Technologies & Tools

• Python (Pandas, Scikit-learn, TensorFlow, stable-baselines3 for reinforcement learning, SHAP/LIME for explainability)
• Grasshopper + Ladybug Tools
• Revit (geometry extraction + BIM data)
• Streamlit (interactive design interface)
• EnergyPlus / OpenStudio (performance simulation)
• Git for version control

🔬 Research Themes

This work is informed by research in: - Reinforcement learning for adaptive generative design optimization - Explainable AI techniques for model transparency and decision support - Generative design for architecture and layout planning - Machine learning models for energy prediction - Early-stage performance-based design - Passivhaus and zero-energy building standards

📖 See docs/literature_review.md for citations and background readings.

📚 Learning Roadmap & Skills

This project is also part of a structured self-learning journey to prepare for PhD-level research. Key learning milestones:

| Stage | Topic | Tools / Books | Outcome | |-------|----------------------------|--------------------------|---------| | 1 | Python Fundamentals | Python Fundamentals | Basic scripting and data types | | 2 | Data Analysis | Python for Data Analysis | Manipulate EPW, CSV data | | 3 | Data Visualisation | Matplotlib / Seaborn | Visualise temp, loads, EUI | | 4 | Machine Learning (Scikit) | Hands-On ML by Géron | Train energy prediction models | | 5 | Generative Design | DEAP, Grasshopper | Optimise layouts with GA | | 6 | Simulation Integration | Ladybug, OpenStudio | Link ML to real energy engines | | 7 | UI / App Interface | Streamlit / Dash | Design explorer prototype |

📁 See project notebooks and docs/dev_log.md for progress.

📁 Repo Structure

```text ml-generative-energy-design/ │ ├── data/ ← Raw & processed datasets (CSV, JSON, EPW) ├── notebooks/ ← Jupyter Notebooks for ML and analysis ├── src/ ← Core Python modules ├── results/ ← Model outputs, charts, and metrics ├── tools/ ← Scripts for Grasshopper/Ladybug/Revit ├── docs/ ← Research logs, literature review, dev notes ├── learning_roadmap/ ← Books, projects, tutorials, learning logs ├── app/ ← Optional UI with Streamlit or Dash │ ├── .gitignore ├── environment.yml ← Conda environment ├── LICENSE ├── README.md ├── CITATION.cff

```

📚 References

See docs/literature_review.md for citations and background readings.

👤 Author

Ross Edward — MSc in Architectural Technology and Energy Performance, Edinburgh Napier University

🧪 Citation

Please cite this repo using the CITATION.cff file if used in publications.