Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On Vessel Location Forecasting and the Effect of Federated Learning

Andreas Tritsarolis, Nikos Pelekis, Konstantina Bereta, Dimitris Zissis, Yannis Theodoridis

TL;DR

This work addresses vessel location forecasting under data privacy constraints by developing Nautilus, an LSTM-based centralized VLF model, and FedNautilus, a federated extension. Nautilus delivers competitive, often superior, short-term predictions (up to 60 minutes) on multiple AIS datasets, while FedNautilus exposes challenges from client drift due to data heterogeneity but gains from personalization (PerFL). The study provides a comprehensive comparison between centralized and federated learning in maritime contexts, including a detailed communication-cost analysis that favors FL in distributed settings. Practical implications include privacy-preserving collaborative VLF with significant reductions in data movement and potential improvements from client-specific personalization. Future work suggests enriching the model with weather and itinerary data and exploring cross-device FL to further optimize the accuracy-cost trade-off.

Abstract

The wide spread of Automatic Identification System (AIS) has motivated several maritime analytics operations. Vessel Location Forecasting (VLF) is one of the most critical operations for maritime awareness. However, accurate VLF is a challenging problem due to the complexity and dynamic nature of maritime traffic conditions. Furthermore, as privacy concerns and restrictions have grown, training data has become increasingly fragmented, resulting in dispersed databases of several isolated data silos among different organizations, which in turn decreases the quality of learning models. In this paper, we propose an efficient VLF solution based on LSTM neural networks, in two variants, namely Nautilus and FedNautilus for the centralized and the federated learning approach, respectively. We also demonstrate the superiority of the centralized approach with respect to current state of the art and discuss the advantages and disadvantages of the federated against the centralized approach.

On Vessel Location Forecasting and the Effect of Federated Learning

TL;DR

This work addresses vessel location forecasting under data privacy constraints by developing Nautilus, an LSTM-based centralized VLF model, and FedNautilus, a federated extension. Nautilus delivers competitive, often superior, short-term predictions (up to 60 minutes) on multiple AIS datasets, while FedNautilus exposes challenges from client drift due to data heterogeneity but gains from personalization (PerFL). The study provides a comprehensive comparison between centralized and federated learning in maritime contexts, including a detailed communication-cost analysis that favors FL in distributed settings. Practical implications include privacy-preserving collaborative VLF with significant reductions in data movement and potential improvements from client-specific personalization. Future work suggests enriching the model with weather and itinerary data and exploring cross-device FL to further optimize the accuracy-cost trade-off.

Abstract

The wide spread of Automatic Identification System (AIS) has motivated several maritime analytics operations. Vessel Location Forecasting (VLF) is one of the most critical operations for maritime awareness. However, accurate VLF is a challenging problem due to the complexity and dynamic nature of maritime traffic conditions. Furthermore, as privacy concerns and restrictions have grown, training data has become increasingly fragmented, resulting in dispersed databases of several isolated data silos among different organizations, which in turn decreases the quality of learning models. In this paper, we propose an efficient VLF solution based on LSTM neural networks, in two variants, namely Nautilus and FedNautilus for the centralized and the federated learning approach, respectively. We also demonstrate the superiority of the centralized approach with respect to current state of the art and discuss the advantages and disadvantages of the federated against the centralized approach.
Paper Structure (14 sections, 2 equations, 7 figures, 4 tables)

This paper contains 14 sections, 2 equations, 7 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Vessel Location Forecasting
  4. Federated Learning
  5. Problem Formulation and Proposed Methodology
  6. Problem Definition
  7. The proposed Nautilus architecture
  8. Extending to FedNautilus
  9. Experimental Study
  10. Experimental Setup, Datasets and Preprocessing
  11. Experimental Results on Nautilus
  12. Experimental Results on FedNautilus
  13. A note on the communication cost
  14. Conclusion

Figures (7)

  • Figure 1: Predicting the future locations of three moving objects (e.g., vessels in the maritime domain) $s_1$, $s_2$ and $s_3$. Given the vessels' current routes up to $t^{s_j}_n$ (solid lines), VLF is used to predict vessels' future locations (green points) at three discrete timestamps, $t^{s_j}_n + \Delta t_{\text{next}}$, $t^{s_j}_n + \Delta t_{\text{next}}'$, and $t^{s_j}_n + \Delta t_{\text{next}}"$, respectively.
  • Figure 2: Client-drift in FedAvg is illustrated for 2 clients with 3 local steps ($N = 2$, $K = 3$). The local updates $y_i$ (in blue) move towards the individual client optima $x_i^\ast$ (orange square). The server updates (in green) move towards $\frac{1}{N} \sum_{i}{x_i^\ast}$ instead of to the true optimum $x^\ast$ (black square; DBLP:conf/icml/KarimireddyKMRS20).
  • Figure 3: The proposed Nautilus architecture.
  • Figure 4: Extending to FedNautilus - Federated Learning Workflow
  • Figure 5: Snapshot of (a) Brest; (b) Piraeus; and (c) Aegean dataset, after the preprocessing phase.
  • ...and 2 more figures

Theorems & Definitions (2)

  • Definition 1
  • Definition 2