Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE credits
Automatic Identification Systems (AIS), situational awareness systems (SAS), and other sensory equipment on vessels have generated vast amounts of data on maritime traffic and navigational patterns. The availability of this spatio-temporal data has attracted significant interest from both the research community and the maritime industry, leading to the development of solutions aimed at improving maritime safety and operational efficiency. As maritime traffic increases, so do collision risks, necessitating data-driven safety measures.
This thesis proposes a scalable approach for storing and processing maritime spatio-temporal (ST) data using BigQuery (BQ), with a focus on cost efficiency and optimal query execution. To demonstrate BQ’s effectiveness in supporting ST analysis for maritime safety, we developed a visualization requirements framework incorporating dimensions and metrics designed to extract valuable insights for fleet safety. We also compared two established visualization tools, Metabase and Holistics, to assess their capabilities in meeting these analytical requirements. Additionally, we explored the use of autoencoder (AE) and XGBoost for predicting closest point of approach (CPA) alarms five minutes in advance, thereby enhancing navigational safety.
Our results indicate that BQ provides robust support for ST analysis, capable of processing large datasets at high speeds. For example, 30 GB of data was ingested in under three minutes. Holistics, with its advanced mapping and scalability features, proved particularly effective for maritime data analysis. Furthermore, our evaluation of AE for dimensionality reduction and XGBoost for CPA alarm prediction, achieving nearly 80% recall, underscores the potential of machine learning in maritime safety applications.
2024. , p. 55