Date-Driven Topology Identification in Power Distribution Systems with Machine Learning

With the increase in demand for quality electricity and the number of end-use consumers, the operation and control of power grids have become more and more complex and challenging. Ensuring acceptable reliability and quality of the electricity supply has become particularly important to every aspect of our electrified economy. Due to the growing deployment of Micro-Phasor Measurement Units (μPMUs) in power distribution grids, an abundance of high-resolution measurements is available that can be harnessed for smarter operation and fault analyses in power distribution networks. Traditional models have revealed limitations on the network topology identification which may occupy manpower and material resources with no guaranty to effectively restore power in a short time period when facing faults and other disruptions. This thesis suggests and implements a machine learning framework that uses the μPMU measurements as inputs and provides a full observation of the network topology in real-time. Specifically, the proposed framework employs a Convolutional Neural Network (CNN) to identify the physical state of the power network at all times. The framework was evaluated on the IEEE 34-Node Test Feeder, where the experiments show that the proposed CNN can achieve a promising performance with high accuracy even when the μPMU measurements contain noises and/or missing entries.

Author

• Li, Yifu

Language

• en

Keyword

• power distribution network
• Convolutional Neural Network (CNN)
• topology identification
• Micro-Phasor Measurement Unit (μPMU)

Date created

• 2020

Type of Work

• Master's Thesis

License

• All rights reserved

Rights statement

• In Copyright

GW Unit

• Electrical Engineering

Degree

• M.S.

Advisor

• Dehghanian, Payman

Committee Member(s)

• Ahmadi, Shahrokh
• Imani, Mahdi

Persistent URL

•  https://scholarspace.library.gwu.edu/etd/tb09j638p