Excellence in research: artificial intelligence assisted high-performance parallel computing for power system optimization

The National Science Foundation Office of Integrative Activities awarded Prairie View A&M University $699,999 on October 1, 2026, under the Integrative Activities program (CFDA 47.083) to develop an artificial intelligence-assisted high-performance parallel computing framework and computational toolbox for power system optimization.

The project will create faster, more reliable, and more adaptable methods for optimal operational decision-making in managing the electric power grid while maintaining its reliability, affordability, and resilience. The framework combines artificial intelligence with high-performance parallel computing to divide, solve, and coordinate large operational problems more effectively. The work will develop intelligent methods that adapt the problem-solving process to changing grid conditions and computing resources while preserving physical feasibility and numerical stability. The project addresses limitations in existing parallel optimization methods that depend on fixed decomposition rules, manually selected algorithm parameters, and static computational task assignments, which can lead to slow convergence, uneven processor workloads, and inconsistent solution quality as system conditions change.

Intellectual merit includes new knowledge on how artificial intelligence can guide problem decomposition, improve convergence and solution quality, and manage computing resources within rigorous optimization methods. Broader impacts include strengthening electric-grid reliability and efficiency, supporting national energy security and sustainable energy future, developing educational modules and hands-on training, engaging students, and sharing results through publications, workshops, software tools, and benchmark datasets.

Performance occurs in Prairie View, Texas, with a period of performance from October 1, 2026, through September 30, 2029.

Abstract

This project aims to create faster, more reliable, and more adaptable methods to make optimal operational decisions in managing the electric power grid and maintain its reliability, affordability, and resilience. The project will bring transformative changes by combining artificial intelligence with high-performance parallel computing so that large operational problems can be divided, solved, and coordinated more effectively. This will be achieved by developing intelligent methods that adapt the problem-solving process to changing grid conditions and computing resources while preserving physical feasibility and numerical stability. The intellectual merit of the project includes new knowledge on how artificial intelligence can guide problem decomposition, improve convergence and solution quality, and manage computing resources within rigorous optimization methods. The broader impacts of the project include strengthening electric-grid reliability and efficiency, supporting national energy security and a sustainable energy future, developing educational modules and hands-on training, engaging students, and sharing results through publications, workshops, software tools, and benchmark datasets. The project will develop an artificial intelligence-assisted high-performance parallel computing framework and computational toolbox for power system optimization. Existing parallel optimization methods can reduce solution time, but their performance often depends on fixed decomposition rules, manually selected algorithm parameters, and static assignments of computational tasks. These limitations can lead to slow convergence, uneven processor workloads, excessive communication, or inconsistent solution quality as system size, network topology, operating conditions, uncertainty, and optimization formulations change. The proposed framework aims to address these challenges through three coordinated research thrusts. First, graph-based artificial intelligence, which represents the grid as a network of connected components, and reinforcement learning, will select how an optimization problem should be partitioned across geographic areas, time periods, operating scenarios, functions, or mathematical structures. Second, learning-assisted coordination will improve initialization, tune algorithm parameters, and guide the treatment of discrete decisions and temporary simplifications of selected constraints while the mathematical optimization software remains responsible for feasibility and convergence. Third, reinforcement learning will dynamically assign computational tasks to available processor cores to balance workloads and reduce communication overhead. The approaches will be evaluated by using diverse electric-grid test systems on multicore workstations and high-performance computing platforms. Performance will be compared with conventional approaches using solution time, convergence, feasibility, solution quality, scalability, robustness, communication cost, and processor utilization. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the foundation's intellectual merit and broader impacts review criteria.- subawards are planned for this award.