Using machine learning for unsupervised maritime waypoint discovery from streaming AIS data

Abstract

Estimating the future position of a deep sea vessel more than 24 hours in advance is a major challenge for Dutch logistics service providers (LSPs). Their unscheduled arrival in ports directly impacts scheduling and waiting times of barges, propagating throughout the entire supply chain network. To help LSPs' planners improve planning operations, we intend to capture the characteristics of maritime routes for a specific region (the North Sea connecting the Netherlands and United Kingdom) in the form of a directed graph, which can be used as a foundation for predicting destination and arrival time of each associated vessel.To create such graph we need an efficient way to extract waypoints for traffic data and this is the problem we will address in this paper.Since LSPs only use publicly available data for arrival estimation, our solution is entirely based on Automatic Identification System (AIS) data. Extracting positional information from AIS, we explore various machine learning approaches to identify clusters. We apply DBSCAN algorithm and show its advantages and disadvantages when used on AIS data. The same process is repeated using meta-heuristics, comparing clustering results generated by a genetic algorithm and by modified ant-colony optimization to those produced by DBSCAN. Finally, we present a hybrid approach and its ability to discover waypoints, highlighting the achieved improvements.To extend the problem, two constraints are added. The first is the requirement to handle large volumes of streaming AIS data on standard PC-based hardware. The second introduces the common situation of "dark areas" in a map due to problems with receiving and transmitting AIS data. The algorithm discovers route waypoints in efficient and effective ways under these constraints.

Maritime pattern extraction and route reconstruction from incomplete AIS data

“…Dobrkovic et al [3] investigate various algorithms for waypoint discovery in the area of the North sea, including DBSCAN, modified ant colony optimization, and a genetic algorithm. They explain that in a real case application algorithms have to cope with clustering maritime data points where density varies.…”

“…To lessen this drawback, and allow future comparisons we propose to use a set of simulated AIS data with enumerated routes. We use the same test used in [3] containing four overlapping routes, two horizontal and two vertical (see Fig. 20).…”

“…Upon close inspection of the missed routes, we found the spread of waypoints to be correct in the large segment of the lane, yet the low number of epochs prevented the algorithm to accurately extract waypoints near the starts, and the ends, hence impacting the score. All tests (1)(2)(3)(4)(5)(6)(7)(8)(9)(10) show that increasing the population size has negligible effect on the overall score, and this parameter is removed from further considerations. With tests 4-9, increasing the number of epochs to 800, and finally to 2000, allowed the algorithm to fit all the waypoints, and correctly identify all the lanes.…”

“…As indicated in [3], long-term predictions require identification of sailing patterns for all vessels in the region of interest. Following the successful completion of that step, it is possible to estimate the final position of an entity by associating its current position, and heading to the known pattern for the area, and extrapolating all possible destinations from the graph.…”

“…To obtain current positions of the vessels we use publicly available Automatic Identification System (AIS) data and zoom in on two specific regions in the Netherlands. The research done by [3] and [4] shows that the basis for pattern identification is waypoint extraction, and, although standard clustering algorithms, such as DBSCAN, can be used for this task, the problem of varying traffic density requires a different approach. In [3], we have shown that a genetic algorithm (GA) can be a viable alternative to other machine learning approaches, and although it is resource intensive, it can deliver accurate results once good criteria for the GA fitness function have been found.…”