Background
From a technical perspective, most analytics platforms utilize one of two approaches to model building performance and estimate the impact of various decarbonization strategies. The first approach is using a data model, which might involve database lookups, typology-based assessments, or regression analysis (or some combination of all three) to try and generate insights from observed patterns in the data. These platforms offer roughly the same level of analytics as one could generate from a spreadsheet tool, and they’re great for getting quick answers to simple questions about energy performance, especially at the scale of a portfolio.
The other approach is using a deterministic model, which involves detailed inputs and complex physics-based calculations. Specialized consultants will collect relevant information about existing building characteristics, feed that as an input into building energy models, and then utilize these “calibrated” energy models to simulate achievable energy and carbon reductions with different potential interventions and upgrades. This approach is great for answering more complex, building-scale questions about energy performance, but the process is notoriously time and cost intensive and the results are heavily influenced by modeler judgement and (usually) informed by nothing more than trial and error.
Our Approach
Under the hood, Carbon Signal uses deterministic energy models to analyze energy use, but we use machine learning to speed up and improve the calibration process. This approach combines the best of both worlds: the ability of deterministic energy models to estimate energy flows, and the ability of machine learning to rapidly scale up the calculations to account for hundreds of thousands of variations across a wide array of possible unknowns. The result is a framework that is not only technically superior but also highly extensible. With any specific strategy in a specific building, for example, Carbon Signal can determine the likelihood of achieving a certain range of energy and carbon reductions, even when there is significant uncertainty in he underlying conditions of the building. As the model gets refined over time, the platform automatically recalculates confidence intervals, shrinking uncertainty bounds by closing in on more constrained design options.
Accuracy
How accurate is Carbon Signal? The short answer is that we consider our analytics to be on par with an ASHRAE Level 2 audit. Traditionally, the audit process is handled by specialized consultants who complete detailed engineering calculations based on in-person surveys of existing conditions and data-collection from several sub-metered energy use components. This approach to determine the likely impact of potential interventions has been used and validated for decades in the building engineering industry. But it also sets you back, best case, several weeks and several thousands of dollars, per building, for every building.
The longer answer is that Carbon Signal is built on the hypothesis that any decision-making framework is better served by rapid-response probabilistic models that yield directionally accurate results instead of highly deterministic models that risk being rendered wholly inaccurate with just a few incorrect assumptions. Directional accuracy means the results point towards the correct set of actions with a quantifiable level of uncertainty. Precision, in this case, is derived from probabilities rather than a single predictive element.
This puts less weight on any single assumption, meaning the results are still useful if that assumption turns out to be wrong, and given the inherent uncertainties associated with building operations, this is arguably a better approach to modeling energy use. Carbon Signal enables stakeholders to quickly identify all high-priority projects across a portfolio, ultimately helping them to identify areas for further targeted investigations, while fully understanding the real-world risks and uncertainties associated with their implementation plans.
Academic Research
While working as consultants to develop decarbonization plans for several large university campuses over a decade ago, our founders recognized that the effort and cost associated with traditional assessment methods simply cannot be replicated at the scale and rate at which the existing building stock needs to be decarbonized. Our approach has evolved over time, from a basic spreadsheet tool to track campus carbon emissions to a fully developed API that can analyze any building, anywhere in the world, using only a basic set of information. This evolution can be traced through various academic papers and articles published in peer-reviewed scientific journals.
A comparison of two modeling approaches for establishing and implementing energy use reduction targets for a university campus
S. Nagpal, C. Reinhart | Energy and Buildings, 2018
This paper compares two different traditional energy model approaches to assess a building portfolio –one spreadsheet based, and the second based on archetype building energy models. This work identified the gaps in current techniques and established the need for new workflows.
A methodology for auto-calibrating urban building energy models using surrogate modeling techniques
S. Nagpal, C. Mueller, A. Aijazi, C. Reinhart | Journal of Building Performance Simulation, 2018
To work towards an ideal model that fills the above gaps, while significantly reducing the computational cost of developing building energy models, this paper proposes a hybrid auto-calibration methodology with rapid-response data-driven approximation techniques.
A framework for using calibrated campus-wide building energy models for continuous planning & greenhouse gas emissions reduction tracking
S. Nagpal, J. Hanson, C. Reinhart | Applied Energy, 2019
This paper presents a framework to expand the point-in-time analysis capabilities of conventional energy modeling platforms and proposes a tool for the development and implementation of portfolio-level continuous energy tracking and planning systems.
Auto-calibrated urban building energy models as continuous planning tools for greenhouse gas emissions management
S Nagpal | Massachusetts Institute of Technology, 2019
All the papers mentioned above were comprehensively combined in a doctoral dissertation that presents workflows to evaluate future carbon scenarios for large building portfolios in the form of a continuous planning tool implemented at the campus of Massachusetts Institute of Technology.
