ORCID Identifier(s)

0009-0008-1752-7016

Graduation Semester and Year

Summer 2025

Language

English

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Industrial Engineering

Department

Industrial and Manufacturing Systems Engineering

First Advisor

Victoria C.P. Chen

Second Advisor

Jay M. Rosenberger

Third Advisor

Bill Corley

Fourth Advisor

Wei-Jen Lee

Abstract

Despite the numerous research studies and interest in the non-intrusive load monitoring (NILM) area to improve energy efficiency, the problem of accurate and precise disaggregation of electrical devices has not been solved yet. The goal of our research is to build a method with a focus on higher accuracy on complex state-based appliances, which most approaches struggle to detect due to their power signal complexity and low consumption. Our approach is NILM with data-driven signatures (DS), with the ability to potentially predict power usage over time that would work great for suitable applications such as demand response, anomaly detection, and many others. We use non-linear programming with piecewise continuous function (NLPwPCF) to model the load signatures of complex appliances, such as washing and drying machines. Then we create the database of the most representative feature signatures based on their durations, power patterns, transitions between states, and seasons, which is used in the NILM model to disaggregate the power signal. Data-driven signatures describe the electrical signal over time at least 50\% more accurately than the signatures estimated in other NILM optimization methods based on the average approach. To solve the NILM problem, we build the model using mixed integer linear programming (MILP) that considers the data-driven signatures from the database, appliances' constraints, and eliminates the most improbable loads.

To prove the practicality and accuracy of our method, we use the unique dataset of low-frequency sample size with modern devices, such as electric vehicles (EV) and solar panels (SP), that influence users' behavior and change the total power signal. We apply the control simulation to create the dataset for testing our method vs. the state-of-the-art methods. NILM with DS outperforms competitive deep learning approaches based on accuracy metrics, disaggregation error of the signal, the ability to catch complex appliances with low consumption and reconstruct all modes with the correct sequence. Our approach detects only real loads, ignoring noise, and avoiding false noise. Moreover, we can accurately detect the low-consuming appliances in the total signal even under SP operation, while the accuracy of state-of-the-art methods decreases with SP inclusion into the total power signal. Furthermore, our method is more efficient as its CPU time is significantly faster.

Keywords

Nonintrusive load monitoring, Mixed integer linear programming, Nonlinear programming, Data-driven signatures, Piecewise functions, Smart grid, Power consumption prediction, Complex state-based appliances, Low frequency power data, Solar

Disciplines

Electrical and Electronics | Operational Research | Power and Energy

License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.