Buildings with solar arrays have atypical energy load curves, so we set Gridium Labs upon testing new data science techniques to create algorithms that can accurately handle solar generation.

Solar is sweeping the nation, but its also adding a new challenge to the science of analyzing meter data. Traditional approaches can’t accurately handle the added complexity of a solar array’s inverse effects on the building’s load curve, but big data analytics, applied mathematics, and machine learning algorithms can.

Enter Gridium Labs, our digital garage where we tinker and refine the algorithms, models, and software code that power our products and reveal insights for the buildings using them.

Accurately calculating energy use forecasts day after day, week after week, is not easy. For one thing, it requires tons of data. Pairing your utility meter data with wet-bulb and dry-bulb temperature for that many meters would be impossible on just one computer.

There are ~35,000 data points every year for a smart meter reading once every 15 minutes.

Gridium addressed this issue by building its big data analytics software in the cloud, which is capable of calculating load curve predictions on hundreds of thousands of meters every day. Gridium buildings receiving Snapmeter service will be familiar with the output of these calculations: Monday morning emails flagging any anomalous energy use from the week prior–such as start-up and shut-down behavior, peaks, off-hours use, and elevated baseloads–and a load curve prediction for the week ahead.

One set of experiments conducted in our digital garage centered on improving the accuracy of the algorithms when faced with the effects of solar arrays (and fuel cells). Depending on the generating capacity of the array, it may completely offset your building’s need for utility-provided electricity late in the afternoon, which is exactly the time when a typical building needs the most energy from the grid. Who knew a solar array’s dark side would be a gentle duck curve?

The Snapmeter graph below visualizes this offsetting effect quite clearly on Saturday and Sunday afternoons. The shaded, elevated off-hours baseload use above 50kW reflects the software’s understanding that weekend afternoon use of 0kW should not be considered part of this building’s expected baseload:

Snapmeter image showing baseload.

Similarly, the early-afternoon ramp down and late-afternoon ramp up in energy use are not considered components of the building’s expected start-up or shut-down sequences (see shaded areas):

Snapmeter graph of startup and shutdown.

These are tricky issues, but if you have a solar building, or are considering solar, you can rely on Gridium to provide reliable advice even though your meter looks a little different than others.

Gridium Labs is also capable of quantifying interesting realities of energy use in buildings, such as the percentage of energy used while an average building is unoccupied compared to the percentage of energy used while it’s “on.” Here’s a hint: the answer may not be what you think…

If you have ideas on energy analytics tests we should run, feel free to let us know.

About Millen Paschich

Millen began his career at Cambridge Associates, trained in finance at SMU, and has an MBA from UCLA. Talk to him about bicycling, business, and green chile burritos.

One reply on “Testing solar algorithms in our digital garage”

  1. Ken Cogliano says:

    We currently use your service for the Mazda buildings in Irvine CA.
    We have solar installed at 1421 Reynolds ave, Irvine and we don’t have a log of how much power the solar system puts out per day/week/month
    We do know the Edison power usage but without knowing and logging the solar output we don’t know the total building consumption.
    We really want to be able to accurately do this, what do you suggest?
    Thanks
    Ken Cogliano

Comments are closed.

You may also be interested in...