Online Optimization of Networked Systems Under Uncertain User Decisions and Responses

This project is focused on critical infrastructures such as power systems, transportation networks, and emerging computing platforms. These systems are evolving into far more dynamic networks due to advancements in technology that enable greater participation and responsiveness from users. For example, in power grids, increasing numbers of customers can now react in real time to prices, incentives, and new service offerings, leading to unpredictable and highly variable power consumption patterns; increased volatility is also emerging due to data centers and electrified heating. These changing dynamics and volatility introduce complex uncertainties into the management of these infrastructures, severely limiting the effectiveness of traditional models and optimization strategies that were designed for more predictable conditions. There is a critical need to develop new methods capable of managing these new forms of uncertainty so that infrastructure operators can maintain reliable, cost-effective, and resilient service. The intellectual merit of this project includes mathematical foundations and algorithms to advance control strategies that are essential for maintaining the resilience and reliability of critical infrastructures. The broader impacts include fostering technological innovation in energy, artificial intelligence, and autonomous systems, as well as contributing to workforce development by providing mentoring, research opportunities, and new course offerings. This project seeks to develop and validate a suite of mathematical and algorithmic innovations for modeling and optimizing complex, networked infrastructures operating in environments with uncertainty. Central to the project is a novel integration of classical feedback control and incentive design with modern advances in online optimization and stochastic modeling that accounts for decision-dependent uncertainty. The technical objectives include: (i) developing real-time optimization algorithms that leverage up-to-date system measurements and stochastic gradient techniques to compute control actions and user incentives; (ii) formulating and solving multi-agent, constrained optimization problems that capture the interplay between multiple service providers competing through incentives for user participation, all while ensuring system-wide reliability and operational security; and (iii) designing adaptive methods that harness real-time or data-driven estimates of how users respond to incentives, so optimization remains robust in the face of persistent uncertainty. The project is grounded in the formalism of time-varying stochastic optimization with decision-dependent data and aims to establish general principles and practical approaches for resilient, data-driven infrastructure management. The outcome will be new strategies and algorithms for power grids and similar systems, positively impacting their reliability, resilience, and economic efficiency. 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.