Utility companies trade electricity the way brokers trade stocks: buy low, sell high. That requires guessing how much to generate, when to buy power from another firm, how a heat wave will boost demand and so forth. Operations researcher Samer Takriti of IBM’s Watson Research Center in Yorktown Heights, NY, hopes a computer model using algorithms from financial markets will reduce the uncertainty and help utilities more efficiently meet demand.
The model crunches weather and historical data about electric loads to predict the hour-by-hour demand for electricity. It also considers the operational costs of a utility’s generators-from big nuclear plants that are cheap to run but slow to start and stop to smaller oil or natural-gas plants that are more expensive to run but are quick to start up-and compiles a list of generator scheduling scenarios that maximize revenue. “You can bankrupt the company on a bad forecast,” says Joel Gilbert, CEO of Tucker, GA-based consultancy Apogee Interactive.
IBM is one of several companies tackling energy forecasting, and “Nobody pretends to have a consensus opinion about what model is better than another model,” says Dan Violette of Summit Blue Consulting in Boulder, CO. Takriti would like to improve his model by enabling the computer to sense when conditions-such as temperature-have changed enough that a new prediction is warranted. “You would like the system to have the intelligence, instead of having people hit the solve again’ button,” says Takriti.
To get there, his model will need more computing power and more sophisticated algorithms-enabling it to handle additional variables such as generators’ maintenance schedules and available capacity on the power grid-that could take another year or two, Takriti says. Ultimately, such models might take the guesswork out of energy trading.
Become an MIT Technology Review Insider for in-depth analysis and unparalleled perspective.Subscribe today