Virtual Energy Storage

Battery-like service from intelligent loads

Project Info

Green sources of energy, solar and wind, have a dark side: their intermittency. To deliver electricity when you need it, unless the sun and wind cooperate, we will need giant batteries. But batteries are expensive.
We are working on an inexpensive alternative to batteries that delivers a battery-like service by using consumer loads - air conditioners, water heaters, etc. - intelligently.
The trick is in the phrase "deliver electricity when you need it".
All power consuming devices have enormous flexibility in when they need power to deliver the expected quality of service (QoS) to the consumer. This flexibility is especially large for thermal "loads" such as air conditioners and water heaters used in our buildings. Small increase and decrease in air conditioning power over the nominal does not lead to perceptible change in indoor climate. Yet, when done intentionally, it is equivalent to a small battery charging and discharging. When a large number of such loads are coordinated through intelligent decision-making software over the Internet, a giant "virtual battery" results. After all, loads in buildings consume 75% of the electricity in the USA, so even a 10% flexibility is huge.
The VES service is obtained with existing equipment, with only an addition of software and communication.
While demand side management has been with power grid operators for a long time, the innovation in the VES concept lies in (i) guaranteed bounds on consumers QoS, and (ii) robust and reliable corodination of a large number of loads with minimal communication.

Smart Buildings of the Future

Reduce energy and environmental footprint, personalize comfort all through intelligent decision making

Project Info

Humans spend most of their lives indoors. Buildings and their indoor climate thus determine the health and comfort of humanity. Modern buildings are responsible for enormous energy use and pollution. In the USA, buildings consume 75% of electricity, 34% of primary energy and account for 33% of CO2 emissions. At the same time, buildings seldom do a good job in keeping everyone comfortable and healthy.
The biggest room for improvement is more intelligent operation of its climate control system. Through dynamic decision making that adapts to outdoor weather and changing occupancy levels in real-time, it is possible to improve the buildings' function while reducing its energy use simultaneously. In addition, it should help absorb the highs and lows of solar and wind energy by intelligently managing its power demand. Another need is resiliency. With severity of weather fluctuations increasing, the system must also be resilient to large disruptions, providing the best possible service at the time of such disruptions.
We are working on developing the smarts behind such smart buildings. The biggest challenge is algorithmic: the system must learn and predict a buildings' and its occupants' behavior from available sensors, and be able to optimize its operations in real-time. All without human supervision, 24-7-365. It must adapt to the specific building as well, as every building is distinct. We also work on developing the IoT solutions needed to support these algorithms.

Learning network structure from data

Machine learning methods to discover interactions

Project Info

Most modern engineering systems are large interconnection of smaller systems. Think of the power grid, the Internet, transportation networks, swarms of UAVs, and the climate control system of large building. Analysis of these systems require knowledge of their graph structure, but such knowledge is hard to come by: these systems have grown organically over decades. In this project we are developing methods to discover structure of the network from measurements collected at various nodes of the graph. The technique relies on sparsity-promoting optimization, specifically, l_1regularization, which selects a sparse solution in an underderdetermined situation.