Back to search

International Hellenic University

Modeling novel data-driven machine learning approaches for traffic congestion & energy forecasting

Abstract

dc:description

The transition to sustainable smart cities (SC), intelligent energy and transportation systems is crucial for addressing the challenges of modern urbanization and climate change. The increasing demand for accurate forecasting in energy systems and traffic networks has led to significant advancements in machine learning (ML) and data-driven approaches like timeseries regression and classification. This thesis explores these two critical domains, focusing on multistep-ahead forecasting methodologies and their applications in energy and traffic systems, exploring novel traffic prediction techniques for filling missing information, but also combining the use of traffic forecasts to predict electric vehicle (EV) charging demands. The research is structured around three main pillars of forecasting—Direct, Recursive, and Sequence-to-Sequence (seq2seq) Forecasting—each of which leverages novel model development and ensemble techniques to enhance prediction accuracy. In the energy domain, this thesis contributes to both short-term and medium-term energy forecasting (EF) by developing advanced ML-based models for energy generation and load. Similarly, in the traffic domain, the research addresses traffic congestion prediction using a classification approach for filling missing information or regression for multistep ahead forecasting. These models incorporate factors such as weather, seasonality, user behavior, and system dynamics. Regarding the new novel models presentation, several new approaches have been implemented, one for each multistep ahead timeseries forecasting strategy. Starting with a comprehensive literature review of forecasting techniques for energy load/generation forecasting (EGF/ELF) and traffic congestion prediction (TCP), this research highlights the challenges and state-of-the-art methodologies in each domain. Furthermore, the thesis continues with a novel preprocessing prediction approach for filling missing information on street traffic level. In parallel, two novel multi-step EF approaches were modeled. A dynamic weighted ensemble method was developed for EGF using recursive strategy and a multi-energy, multi-model strategy (MEMMS) for direct heating, cooling and electricity ELF. Taking into consideration the literature gaps and the models' results developed on direct and recursive multistep EF, this research advances the field by introducing a novel seq2seq decision tree (Seq2seqT3DT), a novel tolerance-aware seq2seq tree model with three-dimensional input, a Decision Tree (DT) designed for multi-step forecasting regression tasks, capable for both acurate TCP and EF. The work concludes in a unified framework for EV charging ELF, developing a new logarithmic zero-inflated (LogZI) ensemble model, combining traffic predictions with user-specific and environmental data to enhance EV grid integration. By developing robust and scalable models, the research demonstrates the potential of data-driven approaches to address real-world challenges in these interconnected fields.The results showcase significant improvements in prediction accuracy, computational efficiency and interoperability, with potential implications for smart grid optimization, urban planning, and sustainable development. This thesis not only bridges the gap between energy and traffic systems but also establishes a foundation for future research in integrated forecasting solutions.

Degree

thesis:*
Grantor dc:publisher
International Hellenic University
Year dc:date
2025

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Mystakidis, Aristeidis
  • Μυστακίδης, Αριστείδης

Subjects

dc:subject × 40

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/60059

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

Mystakidis, Aristeidis; Μυστακίδης, Αριστείδης. Modeling novel data-driven machine learning approaches for traffic congestion & energy forecasting. International Hellenic University, 2025. http://hdl.handle.net/10442/hedi/60059