Deep Learning in Demand Side Management: A Comprehensive Framework for Smart Homes

Download files
Access & Terms of Use
open access
Copyright: Li, Chen
Altmetric
Abstract
The advent of deep learning has elevated machine intelligence to an unprecedented high level. Fundamental concepts, algorithms, and implementations of differentiable programming, including gradient-based measures such as gradient descent and backpropagation, have powered many deep learning algorithms to accomplish millions of tasks in computer vision, signal processing, natural language comprehension, and recommender systems. Demand-side management (DSM) serves as a crucial tactic on the customer side of meters which regulates electricity consumption without hampering the occupant comfort of homeowners. As more residents participate in the energy management program, DSM will further contribute to grid stability protection, economical operation, and carbon emission reduction. However, DSM cannot be implemented effectively without the penetration of smart home technologies that integrate intelligent algorithms into hardware. Resident behaviors being analyzed and comprehended by deep learning algorithms based on sensor-collected human activities data is one typical example of such technology integration. This thesis applies deep learning to DSM and provides a comprehensive framework for smart home management. Firstly, a detailed literature review is conducted on DSM, smart homes, and deep learning. Secondly, the four papers published during the candidate’s Ph.D. career are utilized in lieu of thesis chapters: “A Demand-Side Load Event Detection Algorithm Based on Wide-Deep Neural Networks and Randomized Sparse Backpropagation,” “A Novel High-Performance Deep Learning Framework for Load Recognition: Deep-Shallow Model Based on Fast Backpropagation,” “An Object Surveillance Algorithm Based on Batch-Normalized CNN and Data Augmentation in Smart Home,” “Integrated optimization algorithm: A metaheuristic approach for complicated optimization.” Thirdly, a discussion section is offered to synthesize ideas and key results of the four papers published. Conclusion and directions for future research are provided in the final section of this thesis.
Persistent link to this record
Link to Publisher Version
Link to Open Access Version
Additional Link
Author(s)
Supervisor(s)
Creator(s)
Editor(s)
Translator(s)
Curator(s)
Designer(s)
Arranger(s)
Composer(s)
Recordist(s)
Conference Proceedings Editor(s)
Other Contributor(s)
Corporate/Industry Contributor(s)
Publication Year
2022
Resource Type
Thesis
Degree Type
PhD Doctorate
UNSW Faculty