Lead Performer: Golden Analytics – Falls Church, VA
DOE Total Funding: $138,484
Project Term: July 02, 2018 – March 14, 2019
Buildings rarely perform as designed/simulated and there are numerous tangible benefits if this gap is reconciled. The project will develop a new method for energy model calibration that would be more accurate and less computationally intensive than prevailing methods which use brute force Monte-Carlo search on larger parameter spaces. The new method is based on the identification of physically meaningful coarse-grain parameters (e.g., UA of the entire envelope as opposed to individual UAs for walls, windows, roof, and slab) and physically distinct time period (e.g., early mornings when HVAC systems are off). An important innovation is the use of machine learning to correlate residuals (unaccounted for differences between models and measurements) with building operating states, providing additional information to the user.
Model calibration is an important component of energy-efficiency projects for existing buildings. Calibrated models also support building operation applications. Calibration methods that can achieve good fidelity with reduced computational resources have the potential to increase the use of calibration in these projects and applications. They enable the building energy services and utility services communities to make buildings more energy efficient and provide enhanced reliability and flexibility to the electrical grid.