[Paper Review] Building Model Identification during Regular Operation - Empirical Results and Challenges

Qie Hu, Frauke Oldewurtel+4 more|arXiv (Cornell University)|Mar 22, 2016

Building Energy and Comfort Optimization10 references4 citations

TL;DR

This paper proposes a physics-based, multi-zone building model identified from real-time data during regular building operation, addressing uncertainties in internal heat gains from occupancy and equipment. By dynamically updating internal gain estimates using online temperature measurements, prediction accuracy improves by 36% on average compared to fixed models.

ABSTRACT

The inter-temporal consumption flexibility of commercial buildings can be harnessed to improve the energy efficiency of buildings, or to provide ancillary service to the power grid. To do so, a predictive model of the building's thermal dynamics is required. In this paper, we identify a physics-based model of a multi-purpose commercial building including its heating, ventilation and air conditioning system during regular operation. We present our empirical results and show that large uncertainties in internal heat gains, due to occupancy and equipment, present several challenges in utilizing the building model for long-term prediction. In addition, we show that by learning these uncertain loads online and dynamically updating the building model, prediction accuracy is improved significantly.

Motivation & Objective

• To develop a physics-based, multi-zone building model using data collected during regular building operation, avoiding disruptive excitation experiments.
• To address the challenge of large uncertainties in internal heat gains—primarily from occupancy and equipment—on long-term thermal prediction accuracy.
• To improve prediction performance by dynamically updating internal gain estimates in real time using current temperature measurements.
• To evaluate the model’s performance in long-term prediction and assess its suitability for applications like model predictive control and grid ancillary services.

Proposed method

• Conduct excitation experiments during regular operation to perturb the building and collect data for model identification.
• Use a bilinear, multi-zone thermal model to represent the building’s thermal dynamics, incorporating heat transfer between zones and HVAC system behavior.
• Identify model parameters using data from excitation experiments and weekends, assuming zero internal gains during identification.
• Estimate a fixed internal gains function $ f_{\text{IG}}(\cdot) $ from 8 weeks of occupancy and temperature data.
• Implement an online update strategy where the internal gain estimate at time $ k $ is assumed constant from $ k-1 $, using Equation (10): $ f_{\text{IG}}(k) = f_{\text{IG}}(k-1) $, based on current temperature measurements.
• Apply the updated model to predict 24-hour temperature profiles and validate performance against actual data.

Experimental results

Research questions

• RQ1 How accurately can a physics-based, multi-zone building model be identified using data collected during regular building operation without extensive excitation?
• RQ2 To what extent do uncertain internal gains—especially from occupancy and equipment—limit the long-term prediction accuracy of building thermal models?
• RQ3 Can online, dynamic estimation of internal gains significantly improve prediction accuracy compared to using a fixed internal gains function?
• RQ4 What is the impact of model update frequency and estimation strategy on prediction performance in real-world commercial buildings?

Key findings

• The model with a fixed internal gains function achieves an average RMS prediction error of 0.485°C across all zones over a 24-hour prediction horizon.
• The model with online updated internal gains reduces the average RMS prediction error to 0.308°C, representing a 36% improvement in accuracy.
• Zone Center shows the largest improvement, with error dropping from 0.38°C to 0.16°C, indicating high sensitivity to internal gain uncertainty.
• Zone Northwest exhibits the highest baseline error (0.84°C with fixed gains), highlighting challenges in modeling zones with complex thermal characteristics.
• The prediction error increases significantly with longer prediction horizons, primarily due to unmodeled or uncertain internal gains.
• The dynamic update strategy effectively compensates for time-varying internal gains by assuming short-term constancy, improving robustness without requiring prior knowledge of occupancy patterns.