SHARE
Research Highlight

ExaSGD: Using High-Performance Computing to Operate Decarbonized Resilient Grid

Topic:

The Science
A multi-institutional team, featuring ORNL’s Slaven Peles in a key leadership position, has delivered transformational computational capability for power grid planning and operation within ExaSGD, a subproject of Exascale Computing Project. Along the way, the team has solved several mathematical and computational problems including two-level optimization for optimal power flow analysis over an ensemble of contingencies, fine grain parallel evaluation of irregular network models, and parallel linear solve strategies for extremely ill-conditioned systems. The team delivered security constrained optimal power flow analysis for Texas grid with 10,000 contingencies/weather scenarios running within 15min wall clock time.

The Impact
This capability allows for extensive scenario exploration within real time and is critical for resilient operation of a decarbonized grid. Renewable generation introduces significant new uncertainties and faster dynamics that cannot be managed with existing method and tools.

ExaSGD: Using High-Performance Computing to Operate Decarbonized Resilient Grid CSED ORNL Computational Sciences and Engineering
ExaSGD software stack operation schematics: The two-level optimization algorithm computes stochastic security constrained (SC) alternate current optimal power flow (ACOPF) over an ensemble of contingencies/weather scenarios. ACOPF for each contingency is run on a hardware accelerator (GPU). Primal decomposition approach is used to assemble the global SC ACOPF solution. This contingency analysis enforces security constraints and provides solutions that prevent cascading failures.

PIs/Facility Leads: Chris Oehmen (PNNL), Slaven Peles (ORNL)

Publication for this work: Abhyankar et al. "Evaluation of AC optimal power flow on graphical processing units." Proceedings of 2021 IEEE Power & Energy Society General Meeting (2021); Świrydowicz et al. "Linear solvers for power grid optimization problems: a review of GPU-accelerated linear solvers." Parallel Computing (2021): 102870; Wang at al. "An asynchronous distributed-memory optimization solver for two-stage stochastic programming problems." 2021 20th International Symposium on Parallel and Distributed Computing (ISPDC). IEEE, 2021.