No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Deep Representation Learning-Based Vessel Trajectory Clustering for Situation Awareness in Ship Navigation
Abstract
Vessel trajectory clustering using historical AIS data has been a popular research topic in recent years. However, few studies have investigated applying deep learning techniques. In this study, deep representation learning is investigated for use in clustering historical AIS trajectories to provide insight into navigation patterns to support maritime situation awareness. A recurrent autoencoder and beta-variational recurrent autoencoder are investigated to generate fixed size vector representations of the AIS trajectories. Subsequently, clustering is facilitated by applying the Hierarchical Density-Based Spatial Clustering of Applications with Noise algorithm to the representations. The method was tested on historical AIS data for a region surrounding Tromsø, Norway, with successful results. The results also indicate that the beta-variational recurrent autoencoder was able to generate representations of the AIS trajectories that resulted in more compact clusters.
( Video Presentation: https://www.youtube.com/watch?v=KLQp1zwrvk8&feature=emb_logo )
... Yao et al. [38] investigated clustering AIS trajectories using deep representation learning, where the results indicated that the deep learning approach outperformed non-deep learning based approaches. Murray and Perera [39] expanded this work, where it was found that a variational recurrent autoencoder architecture provided better representations for trajectory clustering. These methods, however, do not provide a method to predict the future trajectory of a selected vessel. ...
... To create these local models, it is suggested to cluster historical ship behavior using a variational recurrent autoencoder, as outlined in [39]. This approach is expanded to add more complexity to the model, resulting in improved clustering performance. ...
... It is, therefore, of interest to develop a framework to generate fixed size representations of the trajectories, such that standard clustering techniques can then be applied to the representations. Murray and Perera [39] suggested to utilize a deep representation learning-based approach to facilitate trajectory representation generation for subsequent clustering. The study argues that RNNs are ideal for such a task, as they are designed to generate representations of multivariate sequences via their hidden states. ...
This study presents a deep learning framework to support regional ship behavior prediction using historical AIS data. The framework is meant to aid in proactive collision avoidance, in order to enhance the safety of maritime transportation systems. In this study, it is suggested to decompose the historical ship behavior in a given geographical region into clusters. Each cluster will contain trajectories with similar behavior characteristics. For each unique cluster, the method generates a local model
to describe the local behavior in the cluster. In this manner, higher fidelity predictions can be facilitated compared to training a model on all available historical behavior. The study suggests to cluster historical trajectories using a variational recurrent autoencoder and the Hierarchical Density-Based Spatial Clustering of Applications with Noise algorithm. The past behavior of a selected vessel is then classified to the most likely clusters of behavior based on the softmax distribution. Each local model consists of a sequence-to-sequence model with attention. When utilizing the deep learning framework, a user inputs the past trajectory of a selected vessel, and the framework outputs the most likely future trajectories. The model was evaluated using a geographical region as a test case, with successful results.
... In the case of this study, it is desirable to cluster historical AIS trajectories as illustrated in Fig. 5. Fig. 5 illustrates a subset of trajectory clusters from the data in Fig. 4. These clusters were discovered by applying the approach in Murray and Perera (2021b). This technique leverages a Variational Recurrent Autoencoder (Fabius and van Amersfoort, 2015) to generate fixed size vector representations of historical AIS trajectories, and subsequently clusters the representations using the Hierarchical Density-Based Clustering of Applications with Noise (HDBSCAN) algorithm (Campello et al., 2013). ...
Autonomous ship technology is developing at a rapid pace, with the aim of facilitating safe ship operations. Collision avoidance is one of the most critical tasks that autonomous ships must
handle. To support the level of safety associated with collision avoidance, this study suggests to provide autonomous ships with the ability to conduct proactive collision avoidance maneuvers.
Proactive collision avoidance entails predicting future encounter situations, such that they can preemptively be avoided. However, any such actions must adhere to relevant navigation rules and
regulations. As such, it is suggested to predict encounter situations far in advance, i.e. prior to risk of collision existing. Any actions can, therefore, be conducted prior to the applicability of the
COLREGs. As such, simple corrective measures, e.g. minor speed and/or heading alterations, can prevent close encounter situations from arising, reducing the overall risk associated with
autonomous ship operations, as well as improving traffic flow. This study suggests to facilitate this ability by emulating the development of situation awareness in ship navigators through
machine learning. By leveraging historical AIS data to serve as artificial navigational experience, long-range trajectory predictions can be facilitated in a similar manner those conducted by
human navigators, where such predictions provide the basis for proactive collision avoidance actions. The development of human situation awareness is, therefore, presented, and relevant
machine learning techniques are discussed to emulate the same mechanisms.
Machine Learning (ML) techniques have gained significant traction as a means of improving the autonomy of marine vehicles over the last few years. This article surveys the recent ML approaches utilised for ship collision avoidance (COLAV) and mission planning. Following an overview of the ever-expanding ML exploitation for maritime vehicles, key topics in the mission planning of ships are outlined. Notable papers with direct and indirect applications to the COLAV subject are technically reviewed and compared. Critiques, challenges, and future directions are also identified. The outcome clearly demonstrates the thriving research in this field, even though commercial marine ships incorporating machine intelligence able to perform autonomously under all operating conditions are still a long way off.
Machine Learning (ML) techniques have gained significant traction as a means of improving the autonomy of marine vehicles over the last few years. This article surveys the recent ML approaches utilised for ship collision avoidance (COLAV) and mission planning. Following an overview of the ever-expanding ML exploitation for maritime vehicles, key topics in the mission planning of ships are outlined. Notable papers with direct and indirect applications to the COLAV subject are technically reviewed and compared. Critiques, challenges, and future directions are also identified. The outcome clearly demonstrates the thriving research in this field, even though commercial marine ships incorporating machine intelligence able to perform autonomously under all operating conditions are still a long way off.
Advances in artificial intelligence are driving the development of intelligent transportation systems, with the purpose of enhancing the safety and efficiency of such systems. One of the most important aspects of maritime safety is effective collision avoidance. In this study, a novel dual linear autoencoder approach is suggested to predict the future trajectory of a selected vessel. Such predictions can serve as a decision support tool to evaluate the future risk of ship collisions. Inspired by generative models, the method suggests to predict the future trajectory of a vessel based on historical AIS data.
Using unsupervised learning to facilitate trajectory clustering and classification, the method utilizes a cluster of historical AIS trajectories to predict the trajectory of a selected vessel. Similar methods predict future states iteratively, where states are dependent upon the prior predictions. The method in this study, however, suggests predicting an entire trajectory, where all states are predicted jointly. Further, the method estimates a latent distribution of the possible future trajectories of the selected vessel. By sampling from this distribution, multiple trajectories predicted. Finally, the uncertainties of predicted vessel positions in relation to the estimated trajectories are also quantified in this study.
High-dimensional data can be converted to low-dimensional codes by training a multilayer neural network with a small central layer to reconstruct high-dimensional input vectors. Gradient descent can be used for fine-tuning the weights in such "autoencoder" networks, but this works well only if the initial weights are close to a good solution. We describe an effective way of initializing the weights that allows deep autoencoder networks to learn low-dimensional codes that work much better than principal components analysis as a tool to reduce the dimensionality of data.
Variational Recurrent Auto-Encoders
- O J R Fabius
- Van Amersfoort
Fabius, O. & J. R. van Amersfoort (2015). Variational Recurrent
Auto-Encoders. In Proceedings of the International Conference on Learning Representations (ICLR).