In addition to forecasting demand, Gridium’s Snapmeter reports provide retrospective analysis of energy use patterns. Often these patterns are easily visible to the naked eye. Energy spikes, equipment cycling, and hard starts stand out sharply from the background noise. Today I’m going to show you how to diagnose subtler issues revealed through Snapmeter’s statistical analysis.
Check out the following energy use and demand values taken from a Snapmeter report (click to enlarge):
Alongside the historical figures, Snapmeter reports variance from expectation, based on a weather-normalized model of normal building behavior. Here, use variance is highlighted in the red box, demand variance in green. As the variance figures make clear, use was extremely high throughout the week.
We can immediately rule out weather as the source of the elevated use, because the variance analysis takes weather into account. We can also see that peak daily demand last week was mostly in line with expectation, except over the weekend, which we’ll come back to in a bit. This is an unusual situation. Peak demand, the single highest daily meter reading, can be quite volatile, driven by equipment problems, weather anomalies, and other temporary conditions. Use, on the other hand, tends to be fairly stable, because fluctuations average out over the course of the day.
Now let’s turn out attention to the last week’s load chart:
The areas highlighted in red reflect high off-hours energy use. Note the solid block of red covering the entire weekend. An uninterrupted 48-hour span of elevated use is unusual in its own right, but here it is made more mysterious by the fact that, to the naked eye, it looks as though the facility manager is doing a stellar job managing weekend use. The weekend load curve is as flat as you could ever hope to see.
Now recall that the variance analysis shows excess peak demand of more than 10% over the weekend. Where is that excess demand in the load curve?
Taken together, these facts point at a likely explanation: the building’s baseload energy use has shifted upwards, settling at a new, higher level. Because Snapmeter’s statistical model will over time adjust to the new baseload, the large variance figures indicate that the shift must have occurred fairly recently. Sure enough, the load chart from the previous week shows a jump from Wednesday to Thursday night.
Zooming in on just Thursday shows the jump more clearly. Baseload is about 30 kW higher than it used to be (the horizontal lines highlight the difference):
30 kW may not sound like a lot, but baseload energy use adds up. Unless corrected, this shift will add about 7% to the building’s electricity bill, or almost $22,000 over the course of a year. All that for a nearly invisible shift in building behavior — invisible, that is, until a statistical model of building behavior shines a spotlight on the unexpected variance.
