Dissertation/Thesis Abstract

Coordinated Voltage Regulation of Distribution Networks
by Li, Changfu, Ph.D., University of California, San Diego, 2020, 142; 28151301
Abstract (Summary)

Electrical power grid is one of the most complex engineering systems. The conventional power grid is designed with centralized power plants and unidirectional electricity flow from plants to consumers on the distribution network. However, there has been increasing adoption of renewable distributed energy resources (DERs) like solar Photovoltaics (PV) generation on the distribution power grid due to their associated environmental and economical benefits. Fluctuating and distributed solar generation can lead to power ramps and two-way power flows, potentially driving the service voltage out of acceptable ranges. Advanced coordinated voltage regulation is needed to integrate high solar penetrations into distribution networks.

In this dissertation, three different coordinated voltage regulation methods are proposed to tackle the voltage regulation challenges arising from increasing solar generation.

The first one optimizes legacy voltage regulation devices (on-load tap changers (OLTCs)). Linearizations are proposed to to reduce computation burden. Comprehensive simulations show that the proposed method enables 67% more PV connection comparing to conventional autonomous OLTC control.

The second one investigates coordination between legacy OLTCs and emerging PV smart inverters with optimization. Nonlinear constraints are relaxed through proposed linearization techniques. Simulations demonstrate improved voltage profiles and significant reduction of OLTC operations from intelligent coordination between OLTCs and SIs. Robustness against forecasting errors (up to 30%) is also validated.

Finally, a data-driven framework using deep reinforcement learning (DRL) is introduced for PV smart inverters coordination. The reward scheme is carefully designed to balance voltage regulation and reactive power generation. Comprehensive tests prove a well-trained DRL agent can achieve near optimal performance with over 99% reduction in computation time comparing to optimization approach.

Indexing (document details)
Advisor: Kleissl, Jan
Commitee: Callafon, Raymond de, Mohsenian-Rad, Hamed, Pawlak, Eugene R., Yu, Nanpeng, Yu, Paul
School: University of California, San Diego
Department: Mechanical and Aerospace Engineering
School Location: United States -- California
Source: DAI-B 82/7(E), Dissertation Abstracts International
Subjects: Electrical engineering, Energy, Mechanical engineering
Keywords: Active distribution network, DER integration, Optimization, Photovoltaic, Reinforcement learning, Voltage control
Publication Number: 28151301
ISBN: 9798557039741
Copyright © 2021 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy