Statistical analysis lets you listen more carefully to what your building is saying
I was recently at a cocktail reception after a busy energy conference. I wandered into a crowd, found some folks I knew and started chatting. It was crowded and noisy, and there were two conversations going on around me at once. I nodded politely, not really understanding either conversation!
Maybe I’m getting old, but as I leaned in and focused on one conversation, the clarity was a relief. I could finally hear and enjoy a conversation.
This is how a good energy model should feel.
Energy data is noisy. Energy models are what allow you to lean in an listen more carefully to what your building is saying.
We previously explained how interval data is changing energy and facility management. Many analyses can be performed using raw data. A simple review of charts and graphs can bring insights that deliver bottom-line savings for your organization. However if you want to really understand your facility, an energy model opens up a whole new layer of insight. Consider the following common facility questions:
- Our September bill was higher than last year. Why?
- We curtailed on Tuesday for demand response. How much load did we shed?
- October 2011 had five weekends; October 2012 has four. How do we compare these bills?
- This summer was really mild. Should I add a little padding to next year’s energy budget? How much?
- How much did that chiller project save us last summer? How much will it save if we have a really hot summer?
- When are we in danger of setting a new demand peak for the current billing cycle?
All these questions can be answered with a good energy model.
At Gridium we think about about two classes of model applications. The first is retrospective and helps explain historical energy performance. This is often called a baseline model. Here, predictive variables such as weather are used to measure energy performance versus expectation. These models are useful for energy reporting, measurement and verification, and operational feedback. The second are predictive models that extrapolate historical behaviors into the future. These models are useful for predicting a peak next week or preparing an energy budget for next year. Good energy management requires both predictive and retrospective models.
Good energy models leverage interval data. Monthly bill data doesn’t provide sufficient granularity to answer the questions above. It’s best to have at least hourly readings for 12 months or more, enough to cover a full cycle of seasons. Good energy models capture not just consumption but also demand, which drives up to 40% of your energy spend. Good models incorporate the major energy drivers, which typically include temperature, humidity, occupancy, baseload, and long-term trends. Good models are not just accurate but also easy to use, which means they don’t burden the user with unnecessary data collection.
Ease of use has traditionally been a deal breaker for energy models. Not many of us have a ready toolkit for performing non-linear regression or other statistical analyses, so, like the drunk searching for his keys beneath the street lamp, we fall back on simpler metrics, even when we know they can mislead. Consider that a typical building’s energy spend will vary 10% from the hottest to mildest year. How do you budget accurately in the face of 10% variance? How do you track progress toward energy reduction goals? How do you measure the ROI from efficiency projects? In energy, a variant of the old mantra applies: you can’t measure what you don’t model.
We’ll cover these issues in more detail in future posts. For now remember that if you want to listen carefully to your building, consider leaning in with an energy model.
