My power company sends me a monthly “Home Energy Report” that purports to show a breakdown of how many dollars’ worth of electricity I used last month by various categories: cooling, kitchen, laundry, lighting, electronics, etc.
Electricity is electricity. My individual appliances aren’t hooked up to communication devices that send data to the power company. When electricity is flowing through the meter, how could they possibly tell whether it was being used for AC, my dishwasher, my clothes washing machine, my TV, or a little bit of each?
I’d be willing to bet that they’re just taking the “standard household” usage pattern- i.e. 30% air conditioning, 10% lighting, 5% television, etc… and figuring out what that pattern looks like for your specific number of kilowatt-hours.
I mean, I suppose they could have some kind of analysis software that can get general ideas of when/how much your AC is running, or your appliances, etc… but I’d think that’s pretty error-prone and vague. And that’s assuming your meter can actually record that sort of thing, even if it is an AMI meter that they can read remotely.
I’m WAGging here, but I suppose they either (1) apply to your aggregate an overall formula according to which the average household uses so-and-so many percent of its consumption for this-or-that purpose, or (2) try to make a guess based on usage patterns. For instance, kitchen appliances typically run in the morning for breakfast and the evening for dinner, the A/C runs typically in the afternoon heat, etc. By observing how much energy you use at what tile of the day, you can guess what purpose it is used for.
But depending on how smart your meter is, they can figure out how much of your load is inductive. Anything with a motor (A/C, laundry, kitchen) is going to contribute to that.
Beyond that, it will depend on how much effort they want to go to. With some effort, a simple model of temperature to load can roughly estimate A/C usage. A sudden kick at odd hours might be laundry. Common usage near meal times will be a dishwasher. And they could compare that with other homes in the neighborhood. Lighting would be correlated with night time hours and square footage and will be primarily resistive (LEDs are still primarily resistive but not entirely the way the old incandescents were). Etc
If they don’t want to go to that much effort, they could use a simple model based on initial study of ‘typical’ usage in their service area.
They can probably model it pretty well if they have narrow enough timeslices on the data.
Most electrical appliances have pretty predictable use patterns. Lights and televisions are generally on in the morning and evening, but (mostly) off during the day and at night. Refrigerators cycle on and off at regular intervals. Air conditioners cycle on and off when it’s hot out, but not when it isn’t.
I bet that some test data and a few simple models could approach 90% accuracy for 90% of homes. Obviously there will be outliers, but the model could also probably categorize a lot of those outliers as “this person is doing something atypical with electricity”.
Then they’re probably wrong, because 95% of the time when I start my dishwasher, it’s around 11 PM.
With all the streaming and DVRing going on, I doubt this assumption provides anything remotely like accuracy these days. I literally never turn my TV on at the time that a show is actually starting a real-time broadcast.
We had an energy consultant come in to tell us that they could infer our usage between refrigeration, lighting, climate control and other by installing one smart meter in front of our utility’s meter. They would do this by looking for “signatures” of the draw.
We tested this against a similar location where we had literally dozens of meters to track usage for every type of equipment.
The numbers were not close.
We decided to go with the hundreds of meters option in a dozen stores (different sizes, equipment sets, age of equipment, etc) and extrapolate from that to our thousands of stores.
I asked my wife, a former utility employee. She gave a series of possible explanations.
The most likely is simply that they’re basing it on models. The industry has spent years studying consumption patterns. They get willing homes to allow their usage to be metered through multiple sensors, with known appliances plugged into them. The patterns would then be applied to similar neighborhoods or similarly-sized homes or whatever info they have.
The OP seemingly hasn’t done this. They may have, however, filled out some sort of household survey when they signed up, maybe just checking off what appliances they have. It’s something you may not remember at all. Why would they need this info? It’s very helpful for determining estimated bills. We put in central air conditioning shortly after we bought our house. For whatever inexplicable reasons, this has never registered. For 35 years. And that means our estimated bills keep being wrong year after year.
And the utility can make some educated guesses based on nothing but usage. Studies have been done on individual appliances that yield specific consumption patterns. Just from the tiny spikes and fluctuations in a day’s worth of usage, the difference between a refrigerator and a water bed can be inferred. In fact, when this was first announced people went ballistic because this meant they “knew when they were having sex” and similar nonsense.
In reality utilities put in great efforts NOT to own accessible info that could be applied to individuals. Some AI system scans thousands of systems simultaneously and assigns numbers to a database. No idea how long they hold on to it. Could anybody look at yours? We didn’t get into that. Some legal entity could probably get a subpoena, because they can do that for absolutely everything. The huge spikes in usage from indoor illegal marijuana farms made them findable, but that’s not quite the same thing.
If you see anything you don’t understand on a bill, contact your company and ask questions. Also, every state has a regulating body over utilities that looks into complaints. Almost certainly, the inclusion of any information on a bill has been the end point of a year’s-long process that the state has approved and even dictated down to the last semicolon. But every utility in every state is different, so that “almost” is important.
While it is almost certainly just based on a simple model, there is a whole-house energy sensor out there that purports to use machine learning to divine which appliances are being turned on and off at any time.
The page notes that, for example, you can detect a motor start by virtue of the initial current surge, which later levels out. An electric dryer runs in cycles that can be detected. It’s not just the wattage–although that is useful too–it’s also the short-timescale behavior between different load types.
I’ve no idea how well it works in practice, but if an expert human were given a high-resolution time series of someone’s electrical use, they absolutely could figure out what every load was. There’s nothing wrong with the idea in principle.
I agree with Exapno_Mapcase that there’s little chance that the utilities are actually doing this. But the tech is out there.
As a law enforcement tool, I’ve heard they’ve been looking at usage patterns for a number of years in an effort to detect marijuana cultivation. Apparently the grow lights consume a lot of power, and the power duty cycle is measured and compared to a standard model.
What I’m not seeing is whether there is a smart meter.
I presume for utilities that charge based on time of day, a smart meter is obviously necessary and it at least tracks use by hour or minute. However, anything more than hourly data for each household may be far too much data to collect.
But generally I don’t see how my utility knows anything more detailed than how much power I used in a month (based on my reading) and the general use pattern for the neighbourhood.
The data could be processed locally. The device I linked to above collects power data at 1 megasample/sec. However, it reduces that to things like “dryer turned on at 10:03, off at 11:12”, which is really a small amount of data even for a low-bandwidth radio link.
I sort of doubt that the meters themselves are set up in such a way that their time slices are small enough to do a whole lot that’s meaningful like that. I mean, if I go look on the Oncor website and look for “smart meter”, it says outright that it looks at 15 minute intervals.
I suspect it’s awfully hard to pick out the signature of a fridge compressor’s inrush current draw and the current it uses to run in a 15 minute cycle, especially if there are other things going on that use a lot of power.
That’s only maybe 4 packets per hour, which even if every single electric user was doing that in a huge city, wouldn’t come close to bogging down any sort of network these days. Even if you quadrupled that, it still probably wouldn’t be a particularly big load.
Most likely a 15 minute interval is granular enough to accurately identify usage patterns in a broad sense for their own usage. They only need a single monthly value for billing purposes, so it’s probably being used by some big AI on their end to model power usage patterns, etc…
Yeah this would be my guess. They need to model power usage throughout the month, and so they have a model that takes into account the outside temperature, the time of day (dinner time, afternoon TV watching, time lights come on, time going to bed etc.) As part of this model they create an energy use profile for each house based on optimal Bayesian weight selection for the different components, and then add them together to get an aggregate. Then since they have this profile on hand, they might as well show off and send it out to the customer.