TR2024-140

Integrating Generative Machine Learning Models and Physics-Based Models for Building Energy Simulation

    •  Vanfretti, L., Laughman, C.R., Chakrabarty, A., "Integrating Generative Machine Learning Models and Physics-Based Models for Building Energy Simulation", American Modelica Conference, October 2024.
      BibTeX TR2024-140 PDF
      • @inproceedings{Vanfretti2024oct,
      • author = {Vanfretti, Luigi and Laughman, Christopher R. and Chakrabarty, Ankush},
      • title = {{Integrating Generative Machine Learning Models and Physics-Based Models for Building Energy Simulation}},
      • booktitle = {American Modelica Conference},
      • year = 2024,
      • month = oct,
      • url = {https://www.merl.com/publications/TR2024-140}
      • }
  • MERL Contacts:
  • Research Areas:

    Machine Learning, Multi-Physical Modeling, Optimization

Abstract:

This paper describes the integration of generative deep learning models for data-driven building energy simulation. The generative models (GMs) are trained to learn distributions of building input signals from data using Python and PyTorch and interfaced with physics-based Modelica models. The developed integration requirements provide background on typical needs that focus on building energy simulation performance. Simulation examples using models from the Buildings library, refactored to receive GM inputs, are presented to illustrate the benefits of the proposed integration approach and how GMs can be used for building energy performance analysis.