Chinwendu Enyioha — Research

Overview

My areas of research can be broadly defined as network science, encompassing limited communication control, network security, distributed algorithms with applications to smart grid. The ubiquitous nature of information and communication technology has enabled connection between a vast number of components in commercial and military applications. This has led to the evolution of massive networked systems where every individual element – embedded with sensing, computation or decision making abilities now plays a role in improving the overall operation, reliability and efficiency of networked systems. Supporting such systems with large numbers of heterogeneous components interacting involves significant communication. An important example of such system is the power grid where a combination of two-way power and information flow enables control and adjustment of the power demand of individual elements to improve the grid stability. Of course, the power grid is only an example, as distributed computation challenges is inherent in large-scale networked systems.

The fundamental question in my research is – how to design communication-efficient algorithms with provable performance for distributed coordination (and decision making) of large-scale systems without compromising the system objectives? Subareas are summarized below.

Distributed Gradient Methods with Limited Communication

Steady interaction of heterogeneous components especially over communication channels with limited bandwidth poses a challenge to efficient coordination. The overarching goals of this project are to: i) develop novel and applicable distributed algorithms with provable performance for distributed resource allocation subject to communications constraints in power networks. ii) investigate trade-off in achievable performance for different communication constraints, and iii) develop mechanisms that will degrade gracefully with reduced communication bandwidth.

Using gradient-based first order methods, we have shown how a system coordinator may quantize and broadcast a single bit of information that encodes the coordination signal to end-users for them to locally make resource allocation decisions and presented convergence guarantees.

Convergence of Limited Communication Gradient Methods
S. Magnusson, C. Enyioha, N. Li, C. Fischione and V. Tarokh
Submitted to the IEEE Transactions on Automatic Control. 2016

Decentralized Power Allocation with One-Way Communication

Though electric power is a resource that users want and expect to have when needed, it is also a resource in which the growth in its peak demand has exceeded the transmission growth by almost 24% every year. Meeting the power needs of users without overloading the system becomes a challenge, especially since users do not necessarily check the generation capacity of system coordinators before consuming power. The questions here are i) how to develop decentralized power allocation techniques using a one-way communication model that guarantee the system capacity is not overloaded, and ii) how to limit the bandwidth used by the system coordinator to broadcast the coordinating (pricing) signal.

To address these we study a dual descent method in decentralized optimization and derive conditions that guarantee feasibility of the aggregated power usage at each iteration of our algorithm so as to avoid blackout. In addition to proving convergence of our algorithm under these conditions, and establish its rate of convergence, we show how to sacrifice a faster convergence rate at the expense of satisfying feasibility of aggregate consumption at each time.

Practical Coding Schemes for Bandwidth Limited One-Way Communication Resource Allocation
S. Magnusson, C. Enyioha, N. Li, C. Fischione, and V. Tarokh.
To appear in Proceedings of the 55th IEEE Conference on Decision and Control (CDC) , Las Vegas, NV, 2016

Renewable Energy and Volatility of Social Welfare

Today, electric utilities are increasingly tasked with meeting renewable portfolio standards, and are looking to generate power from renewable energy sources wind, solar, and geothermal sources. But integrating the power from such renewable sources into the grid can be a formidable challenge, especially because they disrupt the conventional methods for planning the daily operation of the electric grid. Their power generation fluctuates over multiple time horizons, disrupts the second-to-second balance between the total electric supply and demand, forcing the grid operator to adjust its day-ahead, hour-ahead, and real-time operating procedures. Without proper coordination, the reported potential to harness and integrate geographically distributed renewable energy into the grid brings stability of the grid into question.

This project studies the problem of online power allocation to characterize and understand how much of the fluctuation in aggregate social welfare can be explained by the unstable generation. We derive worst-case bounds on fluctuations between an optimal solution and a real-time solution computed by the OD3 algorithm.

On Variability of Renewable Energy and Online Power Allocation.
C. Enyioha, S. Magnusson, K. Heal, N. Li, C. Fischione and V. Tarokh
Submitted to the IEEE Transactions on Power Systems. 2016

Future Research

Details of my short and long-term research plans are in my Research Statement.

Previous Research

Control of Epidemic Processes on Networks

This project considered models of infection propagation in networks and sought to develop optimal strategies for allocating resources to control the spread of an outbreak in a networked population. We proposed a paradigm shift from network heuristics towards a convex framework for contagion control on general networks – including directed networks with positively weighted edges of arbitrary structure – a first in the literature. Distributed strategies were also studied.

Controllability of Large-Scale Networks

Controlling large-scale networks especially with heterogeneous agents is computationally challenging not only because of the sheer size, but also due to the complex nature of interactions the agents have. This project studied the controllability of a network of linear single-integrator agents when the network size goes to infinity. We provided a theoretical justification to the intuition that high degree nodes pose a challenge to network controllability, by presenting a sufficient condition for controllability of a network based on its structure. In summary, we showed that in graph structures with a bounded maximum degree the controllability Gramian remains well-conditioned even as the network size increases

You can see more papers from these projects in my publications.