How do we ensure that the power grid remains stable, efficient, and reliable as we transition to a future dominated by renewable energy? This project tackled that question by developing new tools at the intersection of control theory, optimization, machine learning, and market design—with the ultimate goal of enabling a smarter, safer, and more sustainable electric grid.

Over six years, the project produced fundamental theoretical advances, designed new algorithms and models, and trained the next generation of researchers in systems and data science.

Key Scientific Contributions (Intellectual Merit)

1. Stability Certification with Recurrence: The project introduced a novel, GPU-parallelizable framework for verifying nonlinear system stability using recurrence—a generalization of Lyapunov methods. This allows scalable and non-conservative safety certification in complex systems.

2. Safe Reinforcement Learning: The project developed foundational algorithms for safe learning, including: 

   • Binary Bellman operators for computing safety critics

   • Barrier-based methods for almost-sure safety with minimal exploration

   • Constrained policy learning grounded in dissipativity theory

3. Electricity Market Analysis: By analyzing two-stage electricity markets (day-ahead and real-time), the project revealed hidden inefficiencies in existing market mitigation policies. These findings have direct implications for how energy markets are regulated and operated.

4. Model Reduction and Network Analysis: The team developed spectral clustering methods for reducing large-scale networked systems while preserving their key dynamic behaviors. These tools help simplify power system models, improve understanding of coherence, and inform controller design.

5. Optimization in Machine Learning: The project studied gradient flow in overparameterized linear models, providing theoretical guarantees that help explain how deep networks generalize and converge during training.

6. Applications Across Fields: Research spanned beyond energy, including work on contact tracing, collision avoidance in robotics, storage scheduling, and cyber-physical systems.

Over 45 peer-reviewed publications were produced in leading venues such as IEEE Transactions on Automatic Control, ICML, ACC, CDC, e-Energy, and L4DC.

Education and Human Capital (Broader Impacts) 

• PhD and Postdoc Training: The project supported over 10 PhD students and postdocs  and mentored multiple undergraduates. Graduates have gone on to leading positions at Amazon, the University of Pennsylvania, UC San Diego, and MIT.

• New Course Development: Created and taught: 

  • Mechatronics (undergraduate): Introduced students to embedded control systems and cyber-physical systems.

  • Foundations of Reinforcement Learning (graduate): Integrated control, learning, and decision-making.

   

• Mentorship and Outreach: The PI mentored undergraduates through JHU’s WISE and SABES programs and supported diverse participation in STEM, including advising underrepresented students.

• Research Dissemination: Results were presented at major conferences (CDC, ACC, ICML, L4DC), workshops (GE EDGE, NREL Autonomous Systems), and in over 10 invited talks globally.

   

Summary 

This project has helped reshaped how we think about real-time control, learning, and decision-making in energy systems. It has laid the theoretical groundwork for safe, scalable, and efficient operation of power systems with high renewable penetration—while also building bridges to machine learning, networked systems, and market economics. The project’s broad scientific, educational, and societal contributions will continue to inform both policy and technology development for years to come.

Last Modified: 04/15/2025
Modified by: Enrique Mallada