Skip to main content

CyLab Chronicles

Anthony Rowe on Wireless Sensor Networks for Building Energy Management

posted by Richard Power

NOTE: In this issue of CyLab Chronicles, we are cross-posting a CyLab Seminar Notes from the Partners Portal. Access to CyLab Seminar Series webcasts, and to the full archive of Seminar videos, is an exclusive benefit of membership in the CyLab Partners program. But from time to time, we release individual videos both to highlight the vital nature of CyLab research and to promote the great value of partnering with us.

As part of the CyLab Seminar Series for 2011-2012, Dr. Anthony Rowe spoke on Wireless Sensor Networks for Building Energy Management.

Here are few brief transcribed excerpts from Rowe’s talk. They are meant merely to whet your appetite and encourage you to view the full seminar, which you will find embedded below.

"To get started, I want to motivate the talk by looking at where energy usage typically goes, in the U.S., right now. … Industrial takes about 27% of the energy, and they have been conscious about energy for some time, so there has been a lot of work on optimizing that already. Transportation takes up another 34%, and transportation is a hot topic, people are talking about hybrid cars, public infrastructures ... But if you look at the biggest slice of energy usage in the U.S. the 39% consumed by buildings, it tends to harbor a lot of relatively low hanging fruit, when it comes to what we can do from an efficiency standpoint.

"So what I want to talk about today is how sensor networks can improve energy efficiency. We can accurately account for energy costs and load distributions. Imagine having sensors all over the environment, telling you where the energy is going and what devices are using it. We can look for anomalies in the system. We can look at patterns over time. And we can see if some particular aspect of a building is misbehaving, or performing abnormally compared to what it would normally do, and that would flag a facilities maintenance person to inspect or replace equipment. Imagine a system where you have control as well, so you have both sensing and control over the infrastructure. We can start to do things like shut-off unused loads within the system, to help automate the system. We can start to correlate environmental data, occupancy data, time of day usage, energy usage, key profiles of buildings, to do a next layer of optimization across how the system consumes energy on a day–to-day basis. And we can start to monitor appliance performance. It turns out that over time, big consumers like chillers tend to misbehave and deteriorate over time. So we can actually try to identify and target these as places to optimize.

"So the grand challenge here is to eventually support zero net energy buildings, buildings that over an entire climate cycle could actually collect energy from sources like solar or wind power, and be efficient when those aren’t available, to have a zero budget total …

"In my house, right now, we have a system where we have fifty-six different plug meters and sensor devices collecting data. What can we do with that data? Well, we can start doing things like appliance use classification, occupancy-based waste detection and anomaly detection …

"One thing that we do with environmental sensors linked with electric sensors is that we can actually try to classify appliances based on how people interact with them [i.e., Unsupervised Occupancy-Based Appliance Classification]. We can look for things we call 'Background Appliances' (always on). These are things that are always on, always consuming energy, and they are essentially detached from people’s interaction … If we can identify a load as a background load, then we will know all we can do is anomaly detection (for maintenance) … The next class of devices we look at are 'Passive Appliances' (on for fixed period). These are the type of appliance where the user interacts with them at first and then essentially lets them run on their own, without interaction; the classic example would be the washer and dryer. A motion detector in your laundry room would detect some activity at the start of the device and maybe intermittently, but it essentially starts and runs for a fixed duration and then shuts off. Here, of course, we can do anomaly detection and talk about maintenance, but we can also classify this as a potential load for peak shifting … The last category of appliances we would be able to detect are 'Active Appliances' (on with user): TVs, overhead lights, etc. where you have a very strong correlation between occupancy, motion detectors, audio sensors, and the usage of the appliance. So here we can do things like automatically shut loads off. When we see overhead lights left on when no one is there, we can suggest to the user, 'Hey, you left this light on forty-five percent of the time, when no one is there, this is using up x amount of your energy, why don't you either let the system automatically shut it off or at least be more aware.' We can start to think about doing occupancy-based control.

"Another interesting thing we start to see when we do this correlation between sensor values and the actual appliances are clusters of different devices … What you start to see is different sort of usage and location clusters within building environments [i.e., Sensor Data Clustering]. So, for example, in the kitchen, if there is a sound system in the kitchen it tends to be correlated with the kitchen motion detector and the kitchen overhead light; and again, this is what we can pull in to do that classification where we are trying to figure out what types of appliances are responsible for what … This was all done in a completely unsupervised fashion, so we didn't in any way label where the devices where; the system can learn that over time, by looking at correlation. The next step, once you have these clusters built, and you classify things, is that you can now flag areas where there is potential waste [i.e., Active Appliance Waster Estimation] … and then of course, you can start looking at anomaly detection [i.e., Unsupervised Anomaly Detection] … We would like to see that type of labeling come into a user interface to help guide facility managers, or home owners, to monitor their usage … So sort of the last step you can imagine is taking this to the global level (i.e. Context Sensitive Demand Response), thinking outside of just one building or one home. I talked about how information concerning peak shifting could be pushed up, but imagine at each different layer within the smart grid itself, percolating this information up, using it to build a more precise model of how energy could be used and how it could be shifted around, such that people working on transmission and renewable resources have the types of numbers that they need to optimize their systems.

"So in order to support this type of scenario, and I talked early on a little bit about how we have a backbone, I would like to talk a little bit about Internet-Scale Sensing, and the Sensor Andrew project. And for this, hopefully I can give you a demonstration of Sensor Andrew. But the idea here is that more and more physical information is becoming available in the cyber world, and the question we need to ask is 'How can we utilize this information, and how can we share it, and how can we actually put it into a form that is consumable by developers, so that they are not re-inventing the wheel every time they interface with one of these types of devices?' …

"We have a mobile phone version so that you can actually connect to your house remotely through Sensor Andrew, and look at different appliances, and how much energy they consume …

"If you look at the refrigerator, you can see that the refrigerator consumes more energy when you are cooking, because they room tends to heat up, and the refrigerator runs harder. You can see, for example, if you leave the door open, or the door gets stuck open, the energy consumption will keep running in the background. If you see you left the iron on, you want to be able to turn it off, so being able to support that two-way actuation … This is an interesting example of where anomaly detection makes sense; this is showing a plot of my printer. The printer ran out of toner, and the 'Toner low' light started flashing. While the 'Toner low' light was on, it actually kept the toner cartridge hot the entire time, so in that state of blinking, and not putting the system to sleep, it periodically consumed 600 watts. Luckily, after I had the system in place, I realized that, and I was able to stop it. But I am sure there were times in the past where that light was just on blinking for a couple of weeks, before I ordered a toner cartridge.

"So being able to have visibility to these types of things is a big deal."

You will need to install Microsoft Silverlight to view the video.

See all CyLab Chronicles articles