Richard Hillestad is a senior mathematician at RAND. Others who collaborated with him to study an electric utility company were John Adams, Manuel Carrillo, Dan Relles, and the late Al Williams.
Deregulation of the electric power industry in the United States suffered a few shocks this past summer. In California, the first state where deregulation took effect, record wholesale prices for electricity in San Diego prompted the state legislature to impose a rate freeze, which is still in effect. The freeze forced the state's two biggest utilities, Pacific Gas & Electric and Southern California Edison, to absorb $4 billion in costs that they could no longer pass along to customers. A coalition of 26 municipal utilities in the state has even pushed to "re-regulate" the market until problems in the wholesale market are corrected. Re-regulation would mean returning to the pre-1999 practice of tying electricity prices to each generator's cost of producing the electricity. Other states anticipating deregulation look warily toward California.
Some observers blame the California crisis on a lack of surge capacity--that is, the deregulated utilities simply did not amass an adequate supply of electricity to meet summertime peaks in demand. In this view, the recent problems should subside as companies adjust to the deregulated environment, adding new generators to increase capacity and purchasing additional wholesale power. Indeed, the approach to managing risks in the past has been to amass large margins of standby power.
But the problem lies not just in adding enough capacity for a sunny day. The problem is complicated both by the inefficiencies of some utility operations and by the uncertainties inherent in electricity demand, uncertainties such as a sudden shift in the weather and various social factors. Public utilities can sacrifice efficiency for reliability by holding large reserves. Private utilities, on the other hand, will be able to sacrifice neither efficiency nor reliability.
To remain viable, competitive utilities will need to locate and eliminate inefficiencies. To remain reliable, the utilities will need tools to manage the underlying risks of uncertain demand. By reducing unnecessary costs, the utilities can avoid sending huge bills to customers. And by managing unavoidable risks, private utilities can serve the public interest by decreasing the chances of power outages.
We mapped out the operations of a large electric utility that expects to be deregulated soon and that feels the pressure to adjust to a cost-competitive environment. This utility serves about a third of the population of a major state, operates at a peak capacity similar to that of Southern California Edison, and relies on a typical mix of generators using gas-fired steam, gas turbines, coal, and nuclear power.
We worked with this utility to find ways for it to operate more efficiently and strategically. The methods we developed should be widely applicable to other electric power companies. Our recommendations focused on three areas: (1) controlling costs by accounting for daily wear and tear, (2) predicting a sudden change in the weather 48 hours in advance, and (3) developing a decisionmaking tool to help operators manage the risks of residual uncertainty.
We discovered that daily judgments as to when to turn steam generators on and off were highly sensitive to wear-and-tear costs. These costs can be substantial, because the heat cycles involved in turning the generators on and off cause leaks, corrode the metal, and literally burn up the system. When the costs are underestimated, the planning software schedules more starts and stops than it should. Underestimating the costs also undervalues the importance of buying or selling power on the open market versus starting or stopping your own generators.
 AP/WIDE WORLD PHOTOS/DENIS POROY Nora Whitcotton sits in her air-conditioned apartment in El Cajon, Calif., with her oxygen concentrator in July. She suffers from chronic pulmonary problems. Her electricity bill doubled under California's electricity deregulation, which was supposed to lower prices and set the standard for the nation. Bills doubled for many of San Diego Gas & Electric's 1.2 million customers until the state legislature imposed a rate freeze. |
Despite the importance of wear-and-tear costs, there is considerable uncertainty about what these costs actually are for steam-plant generators, which are the generators used most. The estimated costs caused by firing up and cooling down these generators vary widely across the industry, ranging from $1,000 to tens of thousands of dollars, depending on the generating units involved. Therefore, we set out to quantify the effects of alternative wear-and-tear costs.
As it turned out, the scheduling system was very sensitive even to quite minimal wear-and-tear costs. When the scheduling system included the costs, the scheduling software ordered many fewer starts and stops, and there was much less thrashing of the system. When steam-plant units simply remained on, both the number of starts and the total costs were slashed to less than a third of the usual weekly totals (see Figure 1).
Scheduling with these costs in mind could save as much as $200,000 a week for a utility of this size. The utility used this result as an incentive to sharpen its estimates of the wear-and-tear costs for each of its steam-plant generators. The utility now operates with these costs built into its planning and scheduling process.
The utility already used sophisticated tools to predict changes in temperature and subsequent demand. These tools included computerized programs that use a continuous training process to recognize patterns in temperature and demand as well as social determinants of demand (such as work days, school days, weekends, and holidays). On average, these tools work well. They estimate demand correctly within about 1 percent, as long as the temperature is predicted accurately. However, these tools virtually always miss unusual changes in the weather.
On rare days of some years, the utility experienced an unpredicted temperature drop from well above freezing to well below freezing within eight hours or less. On these days, the utility had to drop customers, because it could not start additional generating units fast enough. The weather had just changed too fast. The utility actually needed 24 to 48 hours to plan for these large changes. Sudden cold fronts are especially problematic because of the difficulty of firing up cold units.
To estimate the likelihood of unusual climatic events 48 hours ahead of time, we worked with a meteorologist to understand the mechanism for the unexpected changes in the weather. For cold weather, the mechanism is the movement of a cold front from distances of 1,000 miles or more away. We then obtained weather measurements for North America for the past 30 years. We determined, statistically, the most likely locations and weather measurements that correlated with sudden temperature drops near the utility 48 hours out. We found that certain locations were good predictors, and the important weather measurements were primarily pressure and temperature differences between those locations and the utility's region. The pressure differences drove the cold temperatures. Secondary measurements were cloud and snow cover, which indicated how much the cold temperature 1,000 miles away might rise on its way to the utility's region.
First used in the winter of 1998, this estimator--called the Adams Index, in honor of its creator, John Adams--predicted three "bad days" from January through March of that year. The utility reacted by turning on extra generating units in anticipation. On two of those days, the cold front did come through; on the third, the cold front stalled just short of the utility's region. Even in this last case, however, the utility was able to sell the extra megawatts generated to a neighboring utility.
The uncertainty of electricity supply stems from the potential for generators to break down. The uncertainty of electricity demand stems from the inability to perfectly predict consumer needs for electricity.
A utility can cope with these uncertainties of supply and demand in a number of ways: (a) ignoring some uncertainties, (b) improving computerized predictions of demand, (c) compensating for errors in computerized predictions by switching on some expensive gas-turbine generators at the last minute, (d) buying or selling electricity hastily on the spot market rather than making longer-term purchases or sales, and (e) scheduling additional slack capacity ("spinning reserve"). All of these coping mechanisms carry cost implications.
As an alternative to these mechanisms, we developed a mathematical model to compute the risk level of any planned schedule of generators. This model could then produce alternative schedules for different sets of assumptions and compare the risks and costs of the alternatives with the original schedule. This approach concedes that some amount of uncertainty is inevitable; however, by comparing the consequences of alternative schedules, this approach at least allows plant operators to compare risks and costs and keep them to a minimum.
To estimate the risk of a planned schedule, we started with historical information about the reliability of each generating unit. This information had been routinely collected by the utility. To update these historical estimates, the utility implemented a reporting system in which plant operators revised the data for any units experiencing problems. With the revised data, we could estimate the probability of failure of each generator at each hour of operation. Depending on which generators were scheduled to be operating at any hour, we could then derive hour-by-hour estimates of the megawatts at risk of being lost to generator failure. In addition to this risk on the supply side of the equation, we accounted for the risk of an unexpectedly high level of demand. We calculated this additional risk based on the utility's historical errors in predicting demand. Together, these two estimates constituted the total risk of a planned schedule.
The risk can be illustrated graphically, as shown in Figure 2, which shows the type of display developed for the utility. As the demand for electricity increases, the reserve capacity decreases. In this example, the reserve capacity is expected to decrease nearly to as low as the minimum reserve required by the utility as a safety margin. Plant operators worry when there is a significant risk of the reserve dropping below the established safety line.
Let's assume a worst-case scenario by subtracting the megawatts at risk during an hour of high demand--say, 5 p.m. At first, we subtract the amount of megawatts at risk at the 90 percent confidence level, which means there is only a 10 percent chance that these megawatts will be lost. With this loss, the available reserve dips below the safety margin. If we subtract the amount of megawatts at risk at the 99 percent confidence level, corresponding to a rare 1 percent chance of loss, the reserve would drop below zero. In the latter case, either customers must be dropped, or additional power must be purchased or generated.
The second part of this process allows dispatchers to compare the risks and costs of alternative schedules. The dispatchers can select a period of interest, schedule different generating units, and view the potential outcomes. In addition to the type of reliability risks shown in Figure 2, the dispatchers can also view the comparative effects on startup costs and marginal costs. The marginal costs are important for deciding whether to generate power or buy it on the open market instead.
Using this system, dispatchers can estimate the cost of keeping an unreliable generator on schedule as well as the cost of turning it off. They can decide more confidently if they should make their own power or buy it elsewhere. And they can reduce the worst possible consequences of generator failure or of erroneous predictions of demand.
We believe these approaches to managing costs and risks can be applied broadly across the electric power industry as it shifts to deregulation. Understanding and controlling costs will become much more important for competitive companies. Reliability certainly will remain a primary commodity demanded by all customers. Therefore, many utilities will need decision-support systems similar to those described here.
Even in a deregulated market of private utility companies, reliable electric power will still be an important benefit to the public. Therefore, for the sake of the public interest, let alone for the sake of efficiency, we believe that methods such as those we developed for this utility could be helpful to the electricity industry in general.