Drivers for modern grid planning and design include the rapidly increasing penetration of distributed energy resources, such as wind and solar. These grid attributes are resulting in rapidly increasing complexity, vastly increased fluctuations in voltage, frequency, and other transmission components, and greater uncertainty in supply and demand. Integrating these factors into computational analyses for accurate grid planning and operations with today’s tools could increase the time to solution by 100 times. Utilities need faster results for real-time assessments and decision making.
Building on advances from DOE high-performance computing and grid modeling programs, a research team led by Argonne National Laboratory and Pacific Northwest National Laboratory recently released a new software framework element—called StructJuMP—for this grid optimization challenge. Demonstrated on a security-constrained optimal power flow problem, this new compact framework was 5-10 times faster to set up while achieving runtime performance comparable to compiled languages such as C++. It also ran 28 times faster over 48 processing cores. This performance is comparable to the scaling of the linear optimization version.
StructJuMP uses new mathematical software technologies to allow both faster prototyping and computation for grid analyses that assess reliability and voltage stability. Based on the success of initial modeling runs, the team expects to achieve their project target of more than 100 times improvement compared to serial versions for about 500-1000 scenario problems. This advancement supports DOE objectives to achieve “a 33 percent decrease in cost of reserve margins, while maintaining reliability” in the increased distributed energy resources and renewable penetration model of the future.
The power grid continues to increase in complexity, in good part by the growing use of distribute