Collaborative Research: MoDL: Graph-Optimized Cellular Connectionism via Artificial Neural Networks for Data-Driven Modeling and Optimization of Complex Systems

This collaborative project between University of North Carolina at Charlotte (UNCC) and Clemson University (Clemson) aims at addressing significant national challenges and needs, namely in the fields of artificial intelligence and clean electric power and energy systems. While growing in popularity and diversity of applications, deep learning (DL) methods nonetheless confront challenges especially for modeling complex systems.

These include lack of robustness, scalability, and composability. The research outcomes of this collaborative project will be: i) mathematical tools for understanding and designing a graph-optimized Cellular Computational Network (CCN) for complex system modeling and optimization; CCN suggests a composable modularity that can divide a large system into small subsystems with corresponding computational cells and ii) empowering the operation of carbon-free electric power distribution systems (EPDSs), with goals of improving energy sustainability (while avoiding climate disasters), energy security, and electricity infrastructure reliability.

Furthermore, this collaborative project will provide unique research training to graduate and undergraduate students in the disciplines of artificial intelligence, machine learning, and power systems engineering at the two institutions. The state-of-the-art smart grid equipment at Real-Time Power and Intelligent Systems Lab at Clemson and high-performance computing systems and AI equipment at Synergistic Human+AI Research lab at UNCC will be used to impact outreach activities to high school students.

Underrepresented minority and women groups will be recruited to participate in the research at the two institutions. Therefore, this project contributes to the creation of a new, diverse workforce knowledgeable in machine learning and AI, smart grid/power system technologies, and renewable energy.

Our approach to address the challenging problem of complex system modeling and optimization constitute a novel blend of interdisciplinary study in statistical learning theory, graph theory, control theory, and optimization theory that will lead to novel dynamic system modeling. The project proposes a principled framework and mathematical validation to 1) automatically infer a graph topology from data, 2) develop multi-resolution graph evaluation for reinforcement learning (RL)-based refinement, 3) provide novel and stable reward function design principle for a continuously evolving CCN model, and thus 4) optimize the voltage profile in an EPDS with distributed energy resources.

Overall, our principled mathematical tools for graph-optimized CCN models will broaden the scope of theory and applications in an electric power distribution system.

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.