UC Berkeley

This dissertation is focused on how the integration of inverter-based resources into our power grids are forcing a re-examination of the fundamental principles upon which we operate and analyze these complex systems. Increased electrification and decarbonization of our energy sources are critical if we are to mitigate the impacts of climate change.

Chapter 1 presents a high-level overview of why inverter-based resources are such a transformative technology for large-scale power systems. We discuss how our current power system operational practices are designed around the inherent physical properties of synchronous machines and how we may need to re-visit some of these practices moving forward.

Chapter 2 considers a model of a battery energy storage system and tries to understand, for the voltage source inverter control scheme considered, the importance of modeling the DC side dynamics on the overall behavior of the inverter. We show that, for the controllers considered in this dissertation, we can model the DC side as a fixed voltage source without any loss in accuracy.

Chapter 3 examines how the cyber-physical attack surface of our power systems is changing with the inclusion of inverter-based resources. These dynamical devices perform active control across timescales with their multi-layered digital control loops, whose interaction through the network can result in unexpected system level dynamics. We consider the case of an adversary seeking to control a power electronic load to induce an instability in the system. We demonstrate how the outer-control loop and the phase-locked loop on inverter-based resources might be most vulnerable to attack and what, if any, signatures might appear at the system level during such an attack.

Chapter 4 considers how me might defend the system against the de-stabilizing attack considered in Chapter 3. We propose an adaptive controller that monitors the integrator state of the phase-locked loop for any abnormal sustained oscillatory behavior. In the event that this is present in the signal, we introduce a small amount of stochastic behavior into the control logic of the inverter to invalidate the model of the system that the adversary used in controller design. We show the performance of the proposed controller through simulation on a small microgrid system.

Chapter 5 explores how we might proactively increase the damping ratio of underdamped system modes by using subspace identification to build a reduced-order model that captures the interaction between a local inverter-based resource and the external grid. We show that, by perturbing the reactive power channel of the inverter, we can build a reduced-order model that we can use for controller design to improve the dynamical response of the inverter. We compare the response of inverter-based resources with and without the proposed damping controller in simulation and show improved dynamical response following the tripping of a transmission line.

Chapter 6 moves from control to simulation of large-scale power electronic dominated power systems and explores how the growing field of scientific machine learning might offer opportunities to accelerate time-domain simulations of these systems. Due to the high control bandwidth of inverter-based resources, and their power rating relative to synchronous machines, the number of differential equations needed to analyze these systems is significantly increasing. We show how echo-state networks, trained at the same timestep as was used by an implicit adaptive ODE solver to solve the system, can rapidly, and accurately, predict the system response for parameters outside the training set. We discuss how this approach could be adapted by system operators today as well as some interesting future research directions.

Finally, this dissertation concludes with a brief summary and directions for future research.