ISO New England: 6 month update

ISO New England: 6 month update
Headquarters: Holyoke, MA
Number of Employees: 530
Visit Company Website

ISO New England is the independent, not-for-profit corporation responsible for overseeing the day-to-day reliable operation of New England’s power generation and transmission system, designing, administering, and monitoring the region’s competitive wholesale electricity markets, and managing comprehensive regional power system planning. The company’s workforce of power system engineers, economists, computer scientists, and other professionals fulfill these three critical responsibilities that together ensure New England has reliable, competitively priced electricity today and into the future.

The ISO serves the six-state region of Connecticut, Maine, Massachusetts, New Hampshire, Rhode Island, and Vermont and is regulated by the Federal Energy Regulatory Commission (FERC).

Evaluation of Robust Unit Commitment

Maintaining the reliability of the region’s power system is one of ISO New England’s most important responsibilities. To make certain New England’s 6.5 million homes and businesses have a dependable supply of electricity, the ISO dispatches generators to start up, vary their output, or power down to meet a constantly changing demand known as “load.” An example of load over a typical day is shown below.

In order to ensure power plants are ready to supply electricity when dispatched, the ISO conducts forecasts and schedules the region’s generators on both a day-ahead and real-time basis. During day-ahead scheduling, the ISO considers variables such as historical electricity use, weather patterns, temperature, and power system conditions and then makes decisions on whether or not to run available power plants—a process known as unit commitment (UC) as depicted below. Much of this electricity is generated by baseload power plants, which are generators that produce enough electricity to cost-effectively satisfy the minimum amount of load on the grid, generally operate continuously, and often require a long time (as much as 24-hours notice) to start up. To account for any differences between the forecasted and actual load, adjustments are made throughout the day in real time to bring on or back down supply.

The Challenge: Reliably Integrating Renewables

New England’s policymakers are seeking to increase the amount of renewable resources available to meet consumers’ energy needs. These resources offer low-emission electrical energy, but their variability poses a unique challenge for the reliable scheduling and operation of the region’s power system. A high penetration of variable resources on the grid increases dispatch uncertainty when a large percentage of these resources might not operate as forecasted because the wind isn’t blowing or the sun isn’t shining. Uncommitted generators with long start-up times would be unavailable to relieve the imbalance caused by a large forecast error, or real-time variation. This is a significant problem for the ISO’s existing deterministic UC model, which is unable to account for high levels of uncertainty.

To accommodate the operating characteristics of a large number of variable resources, the ISO will require an effective UC methodology that can produce commitment decisions capable of delivering power within the uncertainty band shown below. ISO New England has been exploring a new UC solution, known as robust optimization, to find solutions that increase grid reliability in the face of such uncertainty. A preliminary experiment conducted by the ISO’s Business Architecture and Technology team in 2010 showed sizable savings on dispatch costs using robust UC techniques compared to the existing deterministic approach.

Ongoing Work

In April 2012, ISO New England kicked off a collaborative project with statistical and computational scientists at Lawrence Livermore National Laboratory (LLNL) to further study whether robust UC would be a more reliable and economic approach to scheduling resources to generate electricity for the grid than the deterministic UC approach. A representation of a scheduling solution is shown below.  Over the course of one year, the team is using LLNL’s high-performance computing capabilities to run a large number of simulations to analyze and optimize a new unit commitment solution that would mitigate the impact of variability and uncertainty on the power system.


During the first half of this project, the team set up the framework for running millions of simulations for both the robust UC and the deterministic UC approaches using data for representative loads generated by a modified Monte Carlo method. During the second half of the project, the team is running simulations and identifying the impact of various factors—such as the level of conservatism or the forecast error (the size of the blue uncertainty band in the load forecast)—on the economic and operational benefits of robust UC over deterministic UC.

Prior to the hpc4energy incubator

Because of the number of inputs, including generating resources, weather conditions, and conservatism levels, up to 1,600 robust UC configurations could be analyzed over the course of the project—with each configuration taking an average of 30 minutes to compute. Solving this set of robust UC problems on one desktop computer would take 800 hours. By parallelizing the software used to solve the problems (using multiple processors to work on the same tasks), ISO New England and LLNL through its HPC Innovation Center reduced the computation time to just 90 minutes. In this solution, 160 computer nodes, each with 10 cores, were used to complete each robust UC problem concurrently. The ability to run this large number of occurrences at one time significantly improved productivity, enabling the team to conduct more comprehensive evaluations in a much shorter time span.

Studying the UC configurations was just the first step in a series of simulations. To obtain statistically significant results, each UC configuration will be accompanied by 1 million dispatch problems—at an average compute time of 15 seconds per solution. Solving 1 million dispatch problems for 1,600 robust UC configurations (a total of 1.6 billion possibilities) on one desktop computer would take 761 years.

As the project moves toward its completion date of April 2013, the team will compare the results of this data and determine the true benefits of the new UC solution.

The hpc4energy incubator provides access to the capabilities of Lawrence Livermore National Laboratory for selected companies in the energy sector to demonstrate the benefits of incorporating high performance computing (HPC) into technology development.  Part of Livermore’s broader industrial outreach and economic development initiatives, hpc4energy is supported by the HPC Innovation Center on the Livermore Valley Open Campus.