National Technical University of Athens (NTUA)
Edge-optimized deep learning & pattern recognition techniques for non-intrusive load monitoring of energy time series
Abstract
dc:descriptionThe rapid growth in global energy demand, coupled with the pressing need for sustainability, necessitates innovative solutions to improve energy efficiency. While technological advancements have introduced numerous energy-saving systems, their standalone implementation often falls short of delivering meaningful impact. Providing users with feedback on their energy consumption behavior has proven to be a critical step toward fostering sustainable practices. Non-Intrusive Load Monitoring (NILM) stands out as a promising approach for providing users with energy feedback by disaggregating the total household energy consumption, recorded by a central smart meter, into appliance-level insights. This empowers users with actionable information to optimize their energy use. NILM’s potential has been significantly enhanced by advancements in artificial intelligence, the Internet of Things (IoT), and the extensive data generated by widespread smart meter rollouts. Although AI-enabled NILM research is promising, its real-world deployment remains limited due to several critical challenges. The first challenge lies in the availability and representativeness of training datasets. Existing NILM datasets primarily originate from regions like the USA and UK, leaving other regions, such as the Mediterranean, underrepresented. This creates a gap in understanding unique energy consumption behaviors, including the use of appliances like air conditioners and electric water boilers, which constitute a significant share of energy consumption and present considerable potential for demand-side energy flexibility. Developing comprehensive datasets that capture these distinct patterns is crucial for training models that can generalize effectively across diverse scenarios. The second challenge pertains to the high computational demands of deep learning models and the limitations of centralized data processing. Deploying NILM systems often requires significant computational resources, typically relying on cloud infrastructure. This not only raises operational costs but also introduces privacy concerns associated with transferring household energy data to central servers. Finally, the reliance on centralized processing hinders scalability and excludes households with limited connectivity or access to cloud services, restricting the practical applicability of NILM solutions. This thesis addresses these challenges through several key contributions. First, it introduces the development of an interoperable data collection framework that facilitates the creation of the Plegma Dataset, a high-frequency energy dataset specifically designed to capture the unique energy consumption behaviors of underrepresented regions like the Mediterranean. Second, it explores advanced deep neural network (DNN) approaches for NILM and investigates cutting-edge compression techniques to optimize these models for deployment on resource-constrained edge devices. Additionally, the thesis delves into the application of compression techniques for NILM models tailored to Mediterranean energy consumption scenarios, using the Plegma Dataset to validate their effectiveness and potential. By bridging the gap between theoretical innovation and practical application, this research aims to establish NILM as a scalable, efficient, and context-aware solution for advancing global energy efficiency.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Athanasoulias, Sotirios
- Αθανασούλιας, Σωτήριος
Subjects
dc:subject × 14- Μηχανική μάθηση
- Ενέργεια
- Βαθειά μάθηση
- Μη Επεμβατική Παρακολούθηση Φορτίου
- Deep learning
- Energy analysis
- Non-Intrusive Load Monitoring
- Edge Machine Learning
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Αυτοματισμοί και Συστήματα ελέγχου
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Automation and control systems
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58939