Deregulation doesn't have to be chaotic, say two researchers who've
designed a software tool to predict electricity prices.
What do you get when you deregulate electric power, an industry twice the size of the long-distance phone business?
Before you answer, bear in mind that with deregulation 200 separate utilities -- generating power from sources as diverse as
natural gas and hydroelectric, solar radiation and nuclear -- will compete to sell energy to whoever wants it, wherever they
may be, at whatever price is competitive. If you guessed the outcome would be a "vast tangle of technical, economic, and
political problems" (as Fortune magazine has put it), you guessed right. Deregulation of the utility industry is
in its infancy: while the federal government has mandated it, only California has actually implemented the process, and then only in the last year. Industry analysts predict 20 percent to 30 percent lower rates, but getting there will be a rough journey, as utilities figure out what to charge consumers, and as customers, especially large corporations, ponder where to get the best deal.
Help may be on the way, however. For three years, Lilian Wu and Samer Takriti of IBM's Thomas J. Watson Research Center have been working on a mathematical model that will allow power companies to predict supply and demand and find the optimal prices for their products, while helping large customers minimize electricity and fuel bills.
| When storms hit the Midwest
during a heat wave, transmission lines went down, and electric power
that had sold for about $30 a megawatt hour jumped as high as $7,000
an hour before sanity was restored. |
 |
OPEN MARKET FOR ELECTRONS
The idea behind deregulation is simple. In the past, each utility had its own monopoly providing energy to its local region. As a result, not only were electricity prices higher than those of oil or gas, but customers in states like New York and California, where demand was high, could end up paying three times as much for their electricity as those in states like Idaho or Kentucky, where it wasn't. Under deregulation, people will be able to buy from the cheapest provider, because power companies will have equal access to each other's transmission lines.
But in the process, the game changes both for utilities and for consumers. In an open market,
utilities can no longer count on
the business of customers within
their territories. For customers, it becomes increasingly important to negotiate the best contract with their utility. In turn, the utility has to know how to price those contracts. "What do you charge for flexibility?" asks Takriti. "What do you charge for fixed rates? If you're a supplier of electricity, you need to be able to predict the selling price. If it's a very hot day, you want to know how much it would cost to buy electricity from the open market compared to what it costs to generate it. If you're a big industrial
consumer with on-site generating capability, you need to understand on any given day what the best fuel is to use for your needs, whether it's electricity,
natural gas or oil."
What makes the situation so complex is the volatility of the electricity market. Despite deregulation, explains Wu, the market is still very local. While electricity can be delivered by neighboring utilities, losses over transmission lines raise costs the farther the electrons have to travel. Moreover, the balance of supply and demand is instantaneous. "You need your power the moment you turn on the switch," says Wu, "but you can't store electricity the way you can store oil or coal. So supply and demand imbalances occur all the time. If a utility has extra power at eight o'clock, it doesn't do it any good at nine o'clock."
To complicate matters further, says Takriti, the volatility of the prices utilities charge one another has been growing year by year. "Three years ago, the highest price was $50 or $60 per megawatt hour," he says. "Two years ago, it reached $300." Last June, when storms hit the Midwest during a heat wave, transmission lines went down, and electric power that had sold for about $30 a megawatt hour jumped as high as $7,000 an hour before sanity was restored. "Nothing justifies these high prices," Takriti says. "It is market participants panicking in response to shortages."
MODELING THE MADNESS
Wu and Takriti were perfectly placed to make sense of this chaos. Wu had spent more than two decades at IBM doing mathematical modeling of markets, and had served as an adviser to President Clinton on science and technology, as well as having sat on several energy panels. Takriti had come to IBM in 1995 after doing his doctoral thesis at the University of Michigan on problems related to the electric power industry.
Shortly after Takriti arrived, the two began working with a Midwestern utility on a "first-of-a-kind" project to help the utility deal with the new deregulated world. "Our customers were interested in two problems," says Takriti. "First, they needed to forecast the electric load in neighboring regions. If neighboring utilities have relatively high demand, then there is an opportunity to sell power. The second problem was how to use that forecast to predict prices in the region."
Step one was to create the part of the model that would forecast the electricity demand for each region. Much depends on the weather, so Takriti and Wu built a model that takes into account all relevant weather information -- including maximum and minimum temperatures, humidity, cloud cover and wind speed -- and then determines which parameters provide the best forecast. When they were done, they had a model that forecast demand over the next 24 hours to within 6 percent. But that was the easy part.
They then had to forecast electricity prices, which are a function of supply and demand. Supply depends on the capacity
in the market, which itself depends not just on the source that generates the electricity -- which varies with the price of natural gas, coal, nuclear and so on -- but on the time of day. "When you switch a generator on," says Takriti, "it has to remain on for a certain number of hours. In the summer, for example, the load peaks in the afternoon, which results in excess generating capacity for the rest of the day, yielding low evening prices."
Wu and Takriti's model for predicting prices consists of three parts. The first part breaks the year into seasons, the week into weekdays, weekends and holidays, and the days into peak and off-peak times. This allows the model to take into account seasonal and other effects. In winter, for example, consumers use both natural gas and electricity for heating. Because utilities also use natural gas for generating power, a strong correlation forms between electricity prices and natural gas prices. In the summer, however, natural gas use is minimal and there is virtually no correlation. "Electricity prices become a function of humidity and temperature," Wu says.
Next, the model generates a range of potential demand and supply, along with probability distributions within the range. Finally, the problem becomes purely mathematical: the model searches for a schedule for the generating units that minimizes the total cost of meeting the electric load. This is no mean feat. "When you take into account a large number of generators, operating hours and possible scenarios, there can be more than a million decision variables in such problems," says Takriti. "We use 300 to 400 scenarios to represent the uncertain future. The difficult task is to optimize such a large-scale model in a reasonable time. Our algorithm solves such problems in 15 to 20 minutes on a moderately powerful computer, yielding a solution that is within a tenth of a percent of the best one possible."
MODELING COMPETITION
Now, Takriti and Wu are working to refine the predictive power of the model by taking into account the competition and cooperation between neighboring utilities. They are concentrating on a concept known as market power, drawn from game theory. "If I know you need electricity and I am one of just a few utilities that can supply it, I have market power," Wu explains. "Using game theory, I can compute what I should charge in order to take advantage of that fact. In building our model, game theory can also help us predict how one or a few players with market power can affect prices."
While Wu and Takriti are optimistic that this new model will improve forecasts still further -- and even predict oddities like last summer's $7,000 prices in the Midwest -- they have yet to amass enough historical data on a deregulated market to hone the model against reality. "Until we have a mature market, it will be hard to verify such models," says Takriti.
In the meantime, the two IBM researchers are consulting to utility companies, hoping to sharpen their models in the process. They've also been working with John Hutsko, Watson's energy program manager, to help it reduce the $4.3 million the lab spends each year on electricity. Located in Yorktown Heights, New York -- some 200 miles from Lake Ontario, where the power is generated -- Watson is in an area with some of the highest electricity rates in the country. The lab has already seen the benefits
of shopping for power. In 1993, says
Hutsko, IBM negotiated a contract to get its electricity from the New York Power Authority, which is run by the state, rather than from Con Edison, the local utility. "That contract alone saved IBM over $1.1 million a year," says Hutsko. "We have now negotiated a new, seven-year contract that promises to save $300,000 a year on top of that."
When deregulation comes into full effect, a tool such as Takriti's and Wu's model could save even more money. "Companies like IBM consume huge amounts of electricity and are very interested in this idea," says Wu. And as Takriti points out, that interest is bound to grow. "The electric power market is expected to triple or quadruple in size as a result of deregulation. It is a great opportunity for IBM."
Gary Taubes is a freelance writer who lives in Venice, California.
