Democritus University of Thrace (DUTH)
Optimization of energy management and power grid operation using non-intrusive load monitoring techniques
Abstract
dc:descriptionThe scope of this doctoral dissertation focuses on leveraging residential smart meter data for appliance-level disaggregation through non-intrusive load monitoring (NILM) and developing a range of home energy management and grid-support applications. Initially, a scalable and real-time NILM system that exploits turn-on transient responses sampled at 100 Hz is introduced. An event detection algorithm automatically identifies appliance activation events and captures the corresponding transient responses in real time. Appliance classification is performed through a combination of a convolutional neural network and a k-nearest neighbor classifier. The system can incorporate new appliances with only a single recorded turn-on transient, eliminating the need for large labeled datasets and training new models. Its performance is validated on a real-world dataset comprising eight distinct appliance types. Moreover, a holistic NILM-assisted home energy management system (HEMS) is developed to optimize day-ahead residential energy demand from a multi-objective perspective. The system incorporates photovoltaic (PV) panels, a battery energy storage (BES) system, multiple heating and cooling units, an electric vehicle (EV), and other flexible appliances. A NILM module is used to disaggregate the energy consumption of flexible appliances and infer end-user preferences. Using this information, an optimization module coordinates flexible loads, heating units, and the BES system to simultaneously reduce energy costs, alleviate EV range anxiety, minimize stress on the distribution network, and lower CO2 emissions, all while maintaining thermal comfort and respecting end-user habits. The approach is validated on three publicly available datasets to assess generalization under diverse conditions through scenario-based analyses spanning an entire year, with performance benchmarked against baseline strategies. In addition, a methodology is proposed that leverages NILM for real-time grid monitoring and operational support. A deep learning model is employed to detect residential EV charging activity in near real time. In the event of critical network conditions, distribution system operators can utilize these insights to identify households located near affected nodes and request them to shift EV charging to off-peak periods, thereby alleviating the issue. The effectiveness of this approach is evaluated through a test scenario simulating undervoltage and line congestion in a low-voltage test feeder with integrated residential end-users and PV parks. Finally, a fully data-driven framework is proposed that harnesses residential flexibility to support distribution network operation and enhance end-user comfort. The framework adopts a hierarchical architecture operating at both grid and household levels to schedule the day-ahead operation of flexible assets. It combines NILM techniques with forecasting models to predict next-day energy consumption patterns of various assets. At the grid level, dynamic operating envelopes (OEs) are employed to secure network voltages, employing model-free voltage calculations. At the household level, a HEMS schedules and coordinates flexible assets to minimize electricity costs, lower CO2 emissions, enhance thermal comfort, and reduce EV range anxiety, all while adhering to the OEs. The framework is evaluated through a year-long simulation using real smart meter data mapped to a low-voltage test feeder, with performance benchmarked against baseline and topology-based strategies.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Athanasiadis, Christos
- Αθανασιάδης, Χρήστος
Subjects
dc:subject × 24- Απόκριση ζήτησης
- Βαθιά μάθηση
- Διαχείριση ζήτησης
- Μη παρεμβατική παρακολούθηση ηλεκτρικού φορτίου
- Μηχανική μάθηση
- Οικιακή ευελιξία
- Προτιμήσεις τελικού χρήστη
- Σύστημα διαχείρισης οικιακής ενέργειας
- Τεχνητή νοημοσύνη
- Artificial intelligence
- Deep learning
- Demand response
- Demand side management
- End-user preferences
- Home energy management system
- Machine learning
- Non-Intrusive Load Monitoring
- Residential flexibility
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Ηλεκτρική και Ηλεκτρονική μηχανική
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Electrical and Electronic Engineering
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/10442/hedi/60899