Date of Award

Winter 3-2018

Embargo Period


Degree Type


Degree Name

Doctor of Philosophy (PhD)


Electrical and Computer Engineering


Gabriela Hug

Second Advisor

Ozan Tonguz


To integrate large volumes of renewables and use electricity more efficiently, many industrial trials are on-going around the world that aim to realize decentralized or hierarchical control of renewable and distributed energy resources, flexible loads and monitoring devices. As the cost and complexity involved in the centralized communications and control infrastructure may be prohibitive in controlling millions of these distributed energy resources and devices, distributed optimization methods are expected to become much more prevalent in the operation of future electric power systems, as they have the potential to address this challenge and can be applied to various applications such as optimal power ow, state estimation, voltage control, and many others. While many distributed optimization algorithms are developed mathematically, little effort has been reported so far on how these methods should actually be implemented in real-world large-scale systems. The challenges associated with this include identifying how to decompose the overall optimization problem, what communication infrastructures can support the information exchange among subproblems, and whether to coordinate the updates of the subproblems in a synchronous or asynchronous manner. This research is dedicated to developing mathematical tools to address these issues, particularly for solving the non-convex optimal power flow problem. As the first part of this thesis, we develop a partitioning method that defines the boundaries of regions when applying distributed algorithms to a power system. This partitioning method quantifies the computational couplings among the buses and groups the buses with large couplings into one region. Through numerical experiments, we show that the developed spectral partitioning approach is the key to achieving fast convergence of distributed optimization algorithms on large-scale systems. After the partitioning of the system is defined, one needs to determine whether the communications among neighboring regions are supported. Therefore, as the second part of this thesis, we propose models for centralized and distributed communications infrastructures and study the impact of communication delays on the efficiency of distributed optimization algorithms through network simulations. Our findings suggest that the centralized communications infrastructure can be prohibitive for distributed optimization and cost-effective migration paths to a more distributed communications infrastructure are necessary. As the sizes and complexities of subproblems and communication delays are generally heterogeneous, synchronous distributed algorithms can be inefficient as they require waiting for the slowest region in the system. Hence, as the third part of this thesis, we develop an asynchronous distributed optimization method and show its convergence for the considered optimal power flow problem. We further study the impact of parameter tuning, system partitioning and communication delays on the proposed asynchronous method and compare its practical performance with its synchronous counterpart. Simulation results indicate that the asynchronous approach can be more efficient with proper partitioning and parameter settings on large-scale systems. The outcome of this research provides important insights into how existing hardware and software solutions for Energy Management Systems in the power grid can be used or need to be extended for deploying distributed optimization methods, which establishes the interconnection between theoretical studies of distributed algorithms and their practical implementation. As the evolution towards a more distributed control architecture is already taking place in many utility networks, the approaches proposed in this thesis provide important tools and a methodology for adopting distributed optimization in power systems.