Electronic Thesis and Dissertation Repository

Thesis Format

Monograph

Degree

Master of Engineering Science

Program

Electrical and Computer Engineering

Supervisor

Pirathayini, Srikantha

Abstract

Climate change and environmental concerns are instigating widespread changes in modern electricity sectors due to energy policy initiatives and advances in sustainable technologies. To raise awareness of sustainable energy usage and capitalize on advanced metering infrastructure (AMI), a novel deep learning non-intrusive load monitoring (NILM) model is proposed to disaggregate smart meter readings and identify the operation of individual appliances. This model can be used by Electric power utility (EPU) companies and third party entities, and then utilized to perform active or passive consumer power demand management. Although machine learning (ML) algorithms are powerful, these remain vulnerable to adversarial attacks. In this thesis, a novel stealthy black-box attack that targets NILM models is proposed. This work sheds light on both effectiveness and vulnerabilities of ML models in the smart grid context and provides valuable insights for maintaining security especially with increasing proliferation of artificial intelligence in the power system.

Summary for Lay Audience

The rapid increase on numbers of smart meters deployment enables utility companies to design and manage various sustainable power consumption programs, such as time-of-use (TOU) and real-time pricing (RTP). Although these programs reduce the engagement of unsustainable and expensive peak-following generation sources, they do not offer so much guidance to improve efficiency of power usage for residents. In this thesis, an ensemble based deep learning model is designed to disaggregate smart meter reading data to appliance-level, so supplying the insight with better granularity. By analyzing and making sense of the disaggregating data, users can engage more into different power usage programs. One potential problem for machine learning based algorithms is they are vulnerable to adversarial attack, which aims to force models making mistakes by only add some indistinguishable perturbations. A novel stealthy algorithm is designed, which can successfully fool deep learning models. It shed lights on the further study to improve the robustness of machine learning model in power grid area.

Download

Share

COinS