My current apartment building has smart meters that measure and report each unit’s electricity usage. While you can get real time high-resolution usage data if you’re willing to stand in front of the meter, the utility company also provides daily usage data through their website.
I got tired of logging in every day, and the visualizations they provide are a little clunky, so I wrote my own Python script to automatically log in, download the data, and graph it using Plotly. The code is available on GitHub, though it’s specific to my utility company (United Illuminating).
WordPress.com doesn’t allow arbitrary embeds, so below is a static image of the Plotly output. Here’s a live version that’s updated every 24 hours. It has mean lines and trend lines (simple linear regression) by month. I also built an Android app for my phone that fetches a vertical layout of this graph in a WebView.
I went to a talk this past week by Elizabeth Wilson, an associate professor of Energy and Environment Policy at the University of Minnesota. It was a really interesting talk that covered a lot of different topics, including some of the privacy implications (and perceptions) of smart meters.
One of her slides contained the following image, which documents electricity demand in a British household:
I found this fascinating; just a few weeks earlier, I had been discussing with a friend visiting from out-of-town how much usage one could infer from reasonable resolution electricity usage data. I was skeptical about how much we could really learn.
- My day-interval data can probably give you an idea of whether I’m home or not, although it might not be any better than just distinguishing between weekdays and weekends. You also might guess when I turned on the heat for the winter.
- This project uses the rtl-sdr dongle to decode the messages sent on the 900 MHz ISM band by a particular brand of smart meter, and receives data every few minutes. I would ultimately like to do this for my brand of meter, which, AFAICT, is broadcasting on the 2.4 GHz ISM band every two minutes or so.
- The graph above uses minute-interval data, derived not from a smart meter, but from a data logger installed at the customer’s home.
- Proof-of-concept research shows that with two-second interval data, it’s possible to determine what television programs are being viewed (assuming you generate fingerprints of viewable content ahead of time).
The graph above has found its way into presentations, papers, and even anti-smart-meter websites, often as an example of how smart meters can reduce privacy and reveal significant details about personal habits.
Curious, I tracked down the original source of the graph. It comes from a paper published in 1999, which tracked the electricity usage habits of 30 British households. But the methodology section contains some critical details that are always omitted when this graph is bandied about:
Questionnaires were designed for completion at the time the logger was installed in order to identify the primary characteristics of the household and in particular which types of appliance were fitted. Some householders also completed tick-box ‘appliance utilisation’ diaries for one weekday and one weekend day during the logging period; these served to improve our ability to ‘read’ the demand profiles… User behaviour is generally deducible from the database of monitored electricity demand profiles and the knowledge of a surveyed household’s appliance stock… the most apparent classification was by periods of occupation.
In other words, the researchers not only had access to minute-interval data about demand, they also knew what appliances people owned, and in some cases knew when people were using these appliances—all because people volunteered this information. The researchers also logged power factor data, and the internal and external temperatures at the home being monitored.
Note the numerous errors on this Stop Smart Meters website, which includes the graph:
- They call it a “hypothetical graph” (it’s a real graph, from a real research paper)
- They imply it was generated by a current-era smart meter (it was generated by a data logger installed at the consumer’s home, which consisted of a 90’s era Acorn PC, and which measured several other variables)
- They claim that the appliance labels in the graph are based on “signatures” derived from variation in power factors (in fact, the original paper notes that power factors are almost always close to 1, except for the refrigerator and washing machine in spin mode; and they omit that these households answered questionnaires and completed diaries about appliance usage)
Without knowing details about appliance ownership, it would be far harder to determine what is going on in household usage. For example, I suspect there is little way to distinguish between an electric kettle and a toaster based solely on usage, other than to guess that one might make tea before one makes toast.
Certainly, some things will always be obvious (you have a refrigerator, you cook using an oven); but others are less transparent. I have an unusual appliance that pulls 1.5 kW for about 90 seconds a few times a day (it’s not a kettle), and I doubt anyone at the utility company will ever guess what it is.
Consumers should realize that smart meters can benefit them. Apart from lowering costs for the utility company and making them more aware about outages, knowing how much you’re using can help you save money. Where I live, electricity costs about $0.21 per kWh, so savings really add up. Utility companies can help by making sure consumers have access to their data. I wish my company offered higher resolution data, but even the day-interval data has been very informative.