Back to search

University of the Aegean

Complex event recognition and forecasting in streams of data

Abstract

dc:description

Mobility data mining is concerned with the extraction of useful insights and the discovery of common patterns in datasets that describe the behaviour of moving objects. The continuous flow of new records generates open-ended data streams. Detecting interesting patterns in these streams is referred to as Complex Event Recognition. Systems capable of recognizing such activities of interest can help achieve situational awareness, based on the most recent information. Similarly, providing timely warnings and estimates on the occurrence of patterns in the future is called Complex Event Forecasting. Focusing on the maritime use case, analysis on vessel trajectories is of utmost importance, witharound 80% of the world’s trade being carried by sea. Taking advantage of vessel tracking systems can lead to the development of Decision Support Systems that help major stakeholders of the shipping industry. These systems can be utilized in both the planning stage and during operations, for route optimization and activity recognition. The most popular monitoring system of vessel traffic is the self-reporting Automatic Identification System (AIS), that includes positional information, along with vessel and voyage specific features. Several studies have emerged in order to model vessel mobility using AIS data, usually focusing on achieving high accuracy on a specific data mining task. Furthermore, the rise of deep learning architectures has resulted in a plethora of approaches suitable for vessel movement forecasting. These approaches have proven to be quite effective in short-term predictive analytics, but few provide long-term estimations. Moreover, the need of large datasets for the generation of accurate models limits their applicability in more infrequent use cases. Finally, the “black box” nature of deep learning models prevents their incorporation into safety-critical real world applications, due to the lack of clarity in their decision making process. In this work, a framework for transforming huge amounts of AIS data, into a descriptive and compact data model, that can be used for identifying the underlying relationships or patterns in an effective way, is presented. The proposed approach is designed for processing raw AIS data towards both descriptive and predictive analytics. A full preprocessing pipeline is described, to filter erroneous and unnecessary positions, while a trajectory reconstruction technique based on historical movement is proposed, for instances of large temporal gaps in the data. Effective and configurable density maps are generated, visualizing the traffic over an area and period of interest. Moreover, focusing on the vessel mobility between two specific ports, an unsupervised multi-step technique for extracting routes of common movement is introduced. Using the extracted routes, irregular activity can be easily identified on data streams of vessel positions, along with more complex events, like loitering activity and search-and-rescue missions. Finally, the extracted routes are translated into a mobility graph, to provide efficient long-term route forecasting for the vessels. The graph is enhanced with classification models that take advantage of vessel features (like their dimensions) or voyage characteristics (seasonality), to provide the best-suited results for each query. The proposed process is evaluated using a large heterogeneous global dataset of trips from vessels of different types, showcasing significant improvement in accuracy over different state-of-the-art methods. The forecasted trajectories are then combined with nautical charts and areas of interest to produce forecasts of potential upcoming vessel activities, like groundings and fishing.

Degree

thesis:*
Grantor dc:publisher
University of the Aegean
Year dc:date
2024

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Troupiotis-Kapeliaris, Alexandros-Nikolaos
  • Τρουπιώτης-Καπελιάρης, Αλέξανδρος-Νικόλαος

Subjects

dc:subject × 16

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/57267

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

Troupiotis-Kapeliaris, Alexandros-Nikolaos; Τρουπιώτης-Καπελιάρης, Αλέξανδρος-Νικόλαος. Complex event recognition and forecasting in streams of data. University of the Aegean, 2024. http://hdl.handle.net/10442/hedi/57267