The major goal of this project is to develop efficient and distributed cyber-physical systems so as to maximize the potential of improving the reliability and stability of the power grid. The specific objectives are to 1) develop efficient fast-converging first-order and second-order distributed algorithms for reactive power control and flow optimization; 2) develop deadline-constrained distributed data aggregation schemes with efficient routing and scheduling decisions; and 3) enable the optimal deployment of the grid monitoring network as an integrated sensing, communication and computing infrastructure. We have focused on the theme of enabling new reactive power control schemes with the additional intelligence afforded by the cyber-physical infrastructure, which can be potentially transformative to the development of stable and reliable smart power grid.

This multi-disciplinary highly collaborative project produced several key outcomes. They can be summarized across two organizing themes:

1)    distributed first- and second-order distributed algorithms design for optimal power flow optimization and disaster recovery;  a) We were the first researchers to develop a distributed Newton’s method for joint vulnerability analysis and load shedding for power grids under extreme events.  We developed distributed second-order interior-point based algorithm that enjoy a fast quadratic convergence rate. Unlike existing first-order gradient-based algorithms that are not only slow converging, but also constantly violate the constraints, our approach guarantees feasibility at all time. Our results contribute to load shedding and disaster recovery that uses second-order distributed techniques.  b) We also developed distributed first-order algorithms that have much lower complexity and better scalability, and yet enjoy competitive convergence performance compared to the second-order approaches. Specifically, we developed a heavy-ball based joint vulnerability analysis and load shedding framework that offers utility-optimality, queue-stability, fast convergence, and low delays. Our results have shown that the proposed heavy-ball method offers an elegant three-way trade-off relationship between throughput, delay, and convergence. We note that this three-way trade-off insight has not been discovered in the literature thus far. 

2)    scalable design and data-driven adaptive control of large-scale cyber-physical networks.  a) We proposed to use fork and join queueing network models to capture the dynamics of modern parallel processing and information fusion in cyber physical networks. We further devleop a necessary and sufficient condition for a fork-join network of arbitrary size to stay throughput scalable.  b)  We developed a novel Bayesian learning framework integrated with a stochastic gradient method to assist the decision making process in preventive vulnerability maintenance for cyber physical networks. The novelty of our methods is that they overcome challenges in real-world data such as censored observation, data scarcity, etc., and are adaptive to changing environments.

The project produced tens of journal and conference publications, many involving faculty and student participants. The results will be published on http://ise.osu.edu/isefaculty/xia/ECCS.html. The project also provided research opportunities for undergraduate and graduate students, including students from underrepresented groups.

Last Modified: 11/29/2016
Modified by: Cathy Honghui Xia