NREL Partnership Demonstrates Computationally Efficient Analysis of Hybrid Plants
When researchers use high-performance computing (HPC)—to model electric vehicle infrastructure or to test the performance of sustainable technologies, for example—they can see how advanced computing impacts their work with its marked computing speed, efficiency, and performance.
Advanced computing accelerates scientific research at the National Renewable Energy Laboratory (NREL). At NREL's Computational Science Center, the Modeling, Simulation, and Optimization Capability (MSOC) team regularly partners with other NREL researchers to help make advanced computing serve them better. In a recent partnership with the U.S. Department of Energy's National Transmission Planning Study to efficiently and quickly analyze parts of the U.S. transmission system, MSOC teamed with other NREL researchers to increase the efficiency and scalability of software used to simulate off-grid, wind-solar plant hourly performance. NREL's HPC is highly scalable and customizable, making it valuable to researchers needing computing solutions for complex challenges.
An office within the Department of Energy tasked NREL to analyze hydrogen costs with optimal hybrid plant design nationwide. Low-carbon hydrogen presents an attractive accelerant for renewable energy in regions with limited renewable resources, but cost considerations are crucial to plant location and design. Utilizing NREL's open-source Hybrid Optimization and Performance Platform (HOPP), researchers set out to analyze gigawatt-scale, off-grid hybrid plant setups—including wind, solar, and electrolyzer assets—at over 50,000 potential hybrid plant locations throughout the United States.
"Running 50,000 sites without HPC would've required external storage and enhanced security protocols," said NREL's Elenya Grant, a mechanical engineering researcher. "Even with 10 people running these simulations, it would've taken days to complete all the sites. MSOC utilized HPC to speed up the process and parallelize code and tasks."
Analyzing over 50,000 sites requires tremendous computing power, and such a sweep can take substantial computational resources and time. Grant met with Ignas Satkauskas, an applied mathematics researcher with MSOC, to work out an approach to perform this task using HPC.
Satkauskas worked with Grant and her team to understand the requirements and limitations of the project. To calculate hybrid plant performance parameters at any site, HOPP software requires renewable resource data reads and optimization calculations. The local version of HOPP retrieved data using NREL REST API, which was a bottleneck to parallelization of the code on HPC. To avoid this limitation, Satkauskas coded up routines that read data from the HPC filesystem using the reX retrieval tool. Next, to parallelize the computations, Satkauskas used Python MPI protocol implementation to distribute the sites over many HPC cores.
Parallelization enables many calculations to run at once and complex problems to be broken down into smaller, more manageable ones that can be solved at the same time. It enables more efficient execution of code, enhancing computing efficiency and reducing costs. As computing grows and complexifies, so too does the size of computing resources needed to solve key challenges.
The parallel, scalable workflow reduced the 50,000-site sweep runtime from 75 days on a laptop to 42 minutes on 100 nodes (3,000 ranks) of Eagle—NREL's previous HPC system. Rapid sweep times allowed for speedy sweep run debugging and made software available for analysis within days.
More efficient utilization of GreenHEART, the integrated code version that includes HOPP, allowed Grant and her team to explore and analyze cost-reduction strategies in off-grid, renewable-powered electrolysis. Those strategies include site selection with abundant wind resources, complementary wind and solar resources, and optimizing the sizing of wind and solar resources to hone the hybrid plant capacity factor. These strategies correlate with increased hydrogen production and reduced electrolyzer stack replacements, resulting in reduced overall costs of hydrogen.
"With efficient computing solutions, we were able to quickly analyze hybrid plants to demonstrate strategies for better hydrogen production benefitting areas of the United States where renewable energy resources may be limited," Grant said. "HPC also allowed us to observe more scenarios than we were able to previously because of the access to large datasets and higher-fidelity models."
Learn more about the Computational Science Center, explore NREL's wind energy technologies research, access the HOPP GitHub repository, or see how the MSOC team can help your project by viewing their competency list.