Back to search

National Technical University of Athens (NTUA)

Machine learning applications for demand response and renewable forecasting: towards optimal market participation

Abstract

dc:description

The global energy transition, driven by decarbonisation and digitisation, is fundamentally reshaping the electrical grid from a centralised, deterministic system into a decentralised, stochastic network. War, constant geopolitical changes and dearness are also making quite an impact on the stability of the energy sector and electricity prices. Within this context, this dissertation investigates the application of ML and AI to address three critical challenges: demand side management, renewable energy forecasting and strategic energy market participation. The primary motivation is to bridge the gap between technological advancement and societal benefit, that can eventually reduce electricity costs for residential consumers and energy communities. To address consumer heterogeneity, this dissertation introduces robust segmentation methodologies. It develops interpretable clustering frameworks for residential electricity and heating load profiles using well-known algorithms, integrated with explainable AI to validate behavioral patterns. Furthermore, a novel hyperparameter-space clustering methodology is proposed, demonstrating that consumers can be effectively segmented based on the structural configuration of their local forecasting models, thereby reducing data transfers and exposure. Regarding supply-side uncertainty, the thesis presents hybrid forecasting architectures that mitigate the risks associated with wind and solar. For wind power, a model fusing physics-informed theoretical power curves with the temporal fusion transformer is shown to significantly reduce forecast errors compared to standard DL baselines. Additionally, a federated learning framework for photovoltaic forecasting is validated, proving that collaborative training can ensure high accuracy while maintaining strict data privacy for prosumers. Finally, optimal strategies for market integration are formulated. is shown that short training windows can improve day-ahead market price predictions by better capturing recent volatility. This insight supports a comprehensive stochastic optimization framework for energy communities operating as virtual power plants. Coordinating community-scale battery storage through an implicit demand response mechanism, the proposed strategy successfully achieves net-zero grid costs, validating a financially viable model for the democratization of energy assets. Concluding, this dissertation validates that advanced ML and data-driven frameworks can effectively balance operational efficiency with data privacy, offering meaningful tools for the current energy landscape. However, as the sector evolves, a critical paradigm shift from isolated predictive models towards holistic, autonomous intelligence is highlighted. They also lay the necessary groundwork for the next evolutionary leap. As the sector matures, these established predictive capabilities are obliged to converge into holistic models-entities that will ultimately realize the vision of a resilient, self-organizing and fully decarbonised energy ecosystem.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Michalakopoulos, Vasileios
  • Μιχαλακόπουλος, Βασίλειος

Subjects

dc:subject × 26

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/61635

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Michalakopoulos, Vasileios; Μιχαλακόπουλος, Βασίλειος. Machine learning applications for demand response and renewable forecasting: towards optimal market participation. National Technical University of Athens (NTUA), 2026. http://hdl.handle.net/10442/hedi/61635