Argonne National Laboratory

Rescuing the Stranded!

January 24, 2017

Many of us during these long wintry days dream of being stranded on a warm, sunny beach with a palm tree providing just the right amount of shade and gentle breeze. But stranded power presents an entirely different image to power system operators. It means electricity that is wasted or unable to be used.

The problem of stranded power is of particular importance with wind power, because of the intrinsic variability of wind load. Although data center operators try to ensure reliable power, oversupply and transmission congestion can prevent generated power from achieving the desired loads at certain times. Exacerbating this problem is the fact that in the case of oversupply, the price that the power generators can get can be so low or even negative that the power is simply dumped. In 2014, for example, the overall stranded power for the Midwest Independent System Operator system was a whopping 7.7 terawatt hours.

One solution is the use of dispatchable loads – loads that are adjustable and can respond in real time to the needs of the power grid.  Researchers at Argonne National Laboratory, the University of Chicago, the University of Wisconsin – Madison are focusing on a particular type of dispatchable load: intermittent, or cloud-based, computing resources. Such resources can be turned on or shut down according to stranded power availability; they can be deployed within seconds; and the computing costs can be defrayed by providing services not related to energy.

For their experiments, the researchers developed an economic dispatch (ED) model designed to minimize the total dispatch cost. They then tested four cases: (1) the base ED model, (2) the base model with 20 additional 200 MW data centers with inflexible (nondispatchable) loads; (3) the base model with colocated wind farms at each of the 20 additional data centers of case 2; and (4) the base model with 20 additional 200 MW data centers operated as dispatchable loads and positioned to minimize the expected total dispatch cost over multiple wind and load scenarios.

“We ran thousands of computationally intensive experiments, requiring the solution of 128,000 linear programs, 8 mixed-integer programs, and 24 stochastic mixed-integer programs each with over 20-million variables and constraints,” said Kibaek Kim, an assistant computational mathematician in Argonne’s Mathematics and Computer Science Division. “But thanks to the use of parallel solvers and a parallel scripting language, we were able to execute the entire set of experiments on the 310-node Blues computing cluster at Argonne in only 25 hours of wall-clock time with 2,000 cores – far less than the over 5 years that would have been required in serial.”

The results for cases 1–2 showed that significant stranded power exists in power grids and that the systems become more vulnerable to wind power variation as the wind level increases. A surprising result was that colocating data centers at wind farm locations (case 3) has little impact on system cost and in fact can increase stranded power. Case 4 was the most promising. The use of optimized dispatchable loads enables better use of wind generation, reduces stranded power, and decreases system costs.

“With our new approach, dispatchable data center loads can achieve up to 60~80% of full capacity. These results provide data center owners a significant economic incentive to use flexible computing for harnessing stranded power,” said Kim.

A paper based on the study, “Data centers as dispatchable loads to harness stranded power,” by K. Kim, F. Yang, V. M. Zavala, and A. A. Chien, appeared in the IEEE Transactions on Sustainable Energy, 8, no. 1, 2017, pp. 208-218.