Enhancing the Performance of Smart Grids via Applying Machine Learning Methods to Smart Meter Data

• Author / Creator

  Pourjafari, Ebrahim

• Conventional distribution systems need to undergo several updates to be able to meet modern energy requirements. While conventional grids have been able to satisfy the objectives they were designed for, lack of sensors and powerful communication systems, primitive control and management methods, weak protection strategies and inability to integrate Distributed Energy Resources (DERs) are some inadequacies of these grids that need to rethought. Smart grid is a vision to address the shortcomings of conventional grids. To do this, the smart grid takes advantage of an advanced measurement and communication system known as Advanced Metering Infrastructure (AMI) to gather data across the grid. The goal is to leverage measurement data to enhance the operation of the system.

  The objective of this thesis is to use the data provided by AMI to train machine learning models that can enhance the operation and management systems of smart grids. The first work in this thesis tries to improve the solution to the Volt-Var Optimization (VVO) problem by introducing an alternative data-driven approach for the current circuit-based VVO methods. Support Vector Regression (SVR) is used to train the data-driven model. Once the model is built, it is used as the internal model of Model Predictive Control (MPC) to optimally satisfy VVO objectives in a closed-loop control system. Since the model is built using the most recent data measurements, it can capture all features of the system. Therefore, in contrast to its circuit-based counterparts, it does not suffer from outdated circuit models.

  Voltage sag readings of smart meters can be used to find the location of faults. The second work in this thesis tries to respond to a situation in which a limited number of AMI-enabled meters are available. The question is where to install those meters so that the observability of the grid is maximized. If locating faults is concerned, the question is where to install smart meters so that the location of faults can be predicted with the maximum accuracy. To find the best locations for installing smart meters, the proposed method in the second work states the fault locating problem as a classification problem. Then using an optimization procedure based on Simulated Annealing, the proposed method finds the optimal placement for smart meters for which the classification model can locate faults with the maximum accuracy. The classification models used in the second work are the Support Vector Machine (SVM) and Naive Bayes (NB) classifiers.

• Subjects / Keywords

  • machine learning
  • smart grids

• Graduation date

  Fall 2019

• Type of Item

  Thesis

• Degree

  Doctor of Philosophy

• DOI

  https://doi.org/10.7939/r3-d5sb-rd27

• License

  Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.