An MIT startup company will soon release a tool that could profoundly change the way electric utilities operate. Here’s how: When the grid becomes overloaded, the utility sends out a call to reduce demand. In response, selected appliances in millions of households talk among themselves and decide—based on random but coordinated selection—which of them will shut off to achieve the needed cuts. The shutdowns cause little or no inconvenience to residents; they affect only customers who have agreed to participate; and there is no central server storing private information or threatening to crash if the participant base gets too large or a hacker gets too successful. Simulations and early trials show that the response to calls for demand reduction will be fast, accurate, and smooth, even on a grid with a high level of constantly fluctuating solar and wind power generation.
A major impediment to widespread deployment of renewable generators is the inability of today’s power grids to deal with their constantly varying output. Without large-scale storage capability, that output can be as changeable and unpredictable as the sun and the wind. The ability to adjust electricity demand in response to changing supply would ease the situation, and the most promising sources of demand flexibility are the residential and light commercial sectors, with their many energy consumers and energy-using appliances, which combined make up almost half of the electricity usage in the country. (The light commercial sector includes small offices, shops, restaurants, and the like.) Studies suggest that people would be willing to cut back their energy consumption when the grid is short on supply. Consumers in some regions of the United States are already getting “smart meters” installed and control devices put on their appliances. But those early trials are not working well from most people’s perspectives. Electric utilities are getting unreliable information, and changes in demand are not fast enough to help for most of their energy-reduction needs. Meanwhile, consumers are strenuously objecting that their privacy is being invaded.
The problem, according to Jacob Beal PhD ’07, research affiliate in MIT’s Computer Science and Artificial Intelligence Laboratory and scientist at Raytheon BBN Technologies, stems from the current need to connect all consumers to a central controller. Moving all the data to and from a central point takes time; the problem gets worse as more people are connected; and the central controller knows everything about everyone—a setup ripe for central failure and abuse of private information. Beal’s solution is simple: Get rid of the central controller. Instead, capture a little of each person’s preferences and willingness to volunteer on their appliances and then set up a network that enables the millions of appliances to work together to shape demand, taking into account their owners’ recorded flexibility.
Beal has long been fascinated by the problem of how to get large numbers of computers or other devices to work together directly, with no central control. Like others working in “amorphous computing,” he began writing computer procedures, or algorithms, based on principles inspired by self-organizing biological systems such as swarms of bees or the human body, which “works well, even though it’s made of many individually faulty pieces,” he explains.
In 2008, Beal realized that his algorithms could have an important energy-related application: They could provide a novel means of controlling the millions of distributed energy-consuming devices that make up residential demand. Under a seed grant from the MIT Energy Initiative, Beal, then-graduate student Vinayak V. Ranade SB ’09, MEng ’10, and others pursued that idea. Their achievements led to the formation of a startup company called ZOME Energy Networks. Co-founded by entrepreneurs Brad Kayton and Jon Rappaport, ZOME has been further developing, demonstrating, and marketing products based on the MIT algorithms, some of which have already been purchased by utility companies worldwide.
The usual approach to assessing flexibility in consumer demand is to ask people how much money would induce them to change their habits and which of several appliances they value more. While the responses provide fodder for optimization studies and economic calculations, the questions do not make sense to the consumer. Says Beal, “We really want to elicit information from people about convenience—convenience versus citizenship, volunteerism, and community spiritedness.”
Their approach—called ColorPower—involves equipping major energy-consuming appliances (or their outlets) with switches that enable the owner to select colors that mimic the messages conveyed by a stoplight. The owner can deem an appliance green, meaning “I can do without this any time, so turn it off or down whenever you need to.” Examples might be turning off the water heater for five minutes or turning up the air conditioner by a degree or two—changes that an individual occupant is unlikely to notice but which, multiplied by thousands of households, will significantly reduce demand.
The owner could instead choose yellow (adjust this device only if the grid is under strain) or red (turn this one off only to prevent a blackout or other emergency). Finally, a large override switch enables the customer to say, “Not now!” Regardless of the assigned color, the device becomes black—”not an option”—but only temporarily. The goal is to discourage the consumer from switching, say, the air conditioner from yellow to red to make sure it keeps running during a crowded Super Bowl party.
Beal stresses the importance of giving consumers choices that they can relate to. “You want to get their real feelings about their energy use, but you don’t want them to make hard decisions,” he says. “The most important thing is not to annoy the customer.”
The information conveyed by color choice drives the second part of the ColorPower system: the algorithms that enable millions of appliances—regardless of their location—to cooperate to achieve the needed total reduction in demand. Each household has a “demand controller” that talks to all the targeted appliances in the house, constantly receiving updates on the color and status of each device. (I’m yellow, and I’m running.) Every household demand controller uses a distributed aggregation algorithm to continually track overall flexibility on the multi-house network. All operations proceed in a decentralized fashion—except when the grid becomes stressed and a central server sends all participants a call to reduce demand.
When that rallying cry is issued, the collective must respond. “Here’s where the ‘coin flipping’ comes in—the feature that makes the system both random and fair,” explains Beal. Say the utility calls for a 30% reduction in demand. In response, each participating appliance flips a coin to determine whether it should stay on or shut off. To get the needed outcome, the coin is weighted so that it has a 30% chance of coming up heads (shut off) and a 70% chance of coming up tails (stay on). On average, 30% of the devices will shut off.
But there are wrinkles that make the calculation not quite so straightforward. For example, the coin flipping is unlikely to yield the exact outcome intended. If the weighting is 30% heads, maybe 25% will come up heads and be switched off. So the distributed aggregator runs again, reports the 5% shortfall, and the still-running devices flip again, aiming for 5% heads. Individual devices will, of course, vary in the amount of power they consume. But the large number of devices and the constant adjustments enable the system to tolerate that variability.
Another complication is that devices should not turn off and on again very quickly, both to prevent damaging the equipment and to keep from annoying the consumer. So an appliance that has been switched—either off or on—cannot be switched again for a set period of time. Slight variability in that delay time ensures that not all subsequent changes happen at once, which could destabilize the grid.
Computing the exact weighting of the coins to be flipped while taking into account the color status of every device “is where all the magic happens,” says Beal. “You throw a lot of algebra and control analysis at it, turn the crank really hard, and get the right probability.” Every household controller has enough information to perform that computation locally and then communicate the outcome so that all devices have access to aggregated information on the network’s current status. The aggregation of data from all households protects the privacy of each household, and the more households there are, the greater that protection.
Thus, while a centrally controlled system starts to break down as the number of participants expands, the ColorPower system only gets better: It becomes increasingly responsive and accurate, and separating out the behavior of an individual household from the aggregated data gets more and more difficult.
To test the system, Beal created a simulator and performed runs to see how quickly and accurately the millions of “appliances” on the simulated network would respond to calls for demand reduction. The figure above presents sample results from one run. Based on his simulations, Beal concluded that the ColorPower system should be able to shape power demand with a high level of accuracy and a delay time of tens of seconds. Such responsiveness makes the widespread use of renewables a practical option. Assume, for example, a grid with significant solar generation. When the sun goes behind a cloud and the power supply abruptly Student Version of MATLAB drops, within a minute, appliances turn down or off to reduce demand to match the new level of supply. When sunshine returns, those appliances switch on with almost no delay.
ZOME Energy Networks is now rolling out products based on the ColorPower approach. Interestingly, their first product is based on Beal’s simulation software. Released in November 2011, the ZOMEimpact Analytic Simulator is now being used by utilities in the United States, Canada, and India to forecast the effects on their power grids of implementing programs to control demand—a new practice in an industry in which simulation has been used only to explore issues on the supply side. In addition, the utilities are using economic models included in the simulation software to help them design new customer-incentive programs.
The demand control product—called ZOMEbalance—is now being demonstrated, and later in 2012, utilities will be able to buy it and offer it as a service to their customers. A working demo presented at the Consumer Electronics Show in January in Las Vegas was well received. ZOMEbalance puts a tenth as much signaling on the network and is up to 100 times faster (in the aggregate) than other products on the market, all of which rely completely on centralized control. The reduced network traffic saves operational costs associated with the smart meter networks, and the fast response time allows for important new uses of demand control.
Kayton and Beal believe that utility programs based on the ColorPower system will appeal to consumers. When the program is available, customers can choose to sign up. After the necessary hardware is installed (about an hour-long job), customers will experience little follow-up other than small but measurable rebates on their bills. While just a few percent of customers have signed up for currently offered demand- management programs, Kayton hopes that with ZOMEbalance the opt-in rate will reach 20% or more. “After all, if you didn’t notice it, you got a benefit, you helped the environment, and it was fair, private, and secure—why wouldn’t you sign up?” says Kayton.
This research was supported initially by a seed grant from the MIT Energy Initiative. Subsequent work is ongoing at ZOME Energy Networks, with support from its investors. Further information can be found in:
V. Ranade and J. Beal. “Distributed control for small customer energy demand management.” Proceedings of the 2010 Fourth IEEE International Conference on Self-Adaptive and Self-Organizing Systems, pages 11–20.
V. Ranade. Model and Control for Cooperative Energy Management. MEng thesis, MIT Department of Electrical Engineering and Computer Science, May 2010.
This article appears in the Spring 2012 issue of Energy Futures.
Press inquiries: firstname.lastname@example.org