Great to have you in Maritime Optima.
You have been working part-time with the company since we founded it in 2018, and during the spring (2021), you will be writing a Master's thesis by using the public data we have access to in Maritime Optima.
When the company first started up in the summer of 2018, I was involved in the project's initial planning and research stages. However, I was working as a summer intern for the Norwegian meteorological institute (MET) at the time, so it was in an unofficial capacity. The following summer, we started mapping and developing a product. During the summer, I created a map data processing system (maps baker) as a foundation for a routing feature, i.e., finding the shortest route between shipping ports and positions.
At the end of the summer, I started working with weather data toward visualizing weather on the map. I continued working with weather data, and the weather visualization feature for the map as a part-time employee after starting the Master in Applied Computer Science (MACS) studies at NTNU the following school year.
During the summer of 2020, I was full-time and a part of the development team working on different application features. I continued being part-time throughout the autumn of 2020.
I have learned a lot working for Maritime Optima; however, the most important I have learned is working as a part of a development team working together on a large and complex project. It has been great working with other students and young developers, and experienced people. I feel fortunate to have been able to be part of a startup environment, and I have learned a lot.
While working for Maritime Optima, I have gained interest in a fascinating industry with many "logistics challenges." There is a digital wind blowing across the maritime industry, and there are so many exciting "data problems" to look into for a developer. I also wanted to work with Maritime Optima throughout the thesis, so I started investigating issues relevant to the company and the industry that were interesting to me. I found many exciting challenges within logistics, digitization, and data analysis. However, I found the most exciting topics during our many discussions using historical AIS data. Maritime Optima also suggested the problem of predicting a vessel's future destination as an exciting challenge. While discovering this had not thoroughly been researched at a larger scale, I decided to look deeper into this problem.
The working title of the thesis is "Vessel destination forecasting based on historical AIS."
The thesis's intended scope is to establish a machine learning (ML) approach to predict vessels' future destinations (ports) by using historical trajectories for similar vessels derived from AIS data. Then, apply the method on a global set of vessels to forecast a global vessel supply image for ports and regions.
Data analysis based on historical AIS data is not a new topic, and it has been explored to a rather large extent. Literature reviews I conducted before starting work on the Master's thesis revealed several motivations behind such work. Some work focuses on predicting vessels' future positions in a limited time interval and geographical scale that has, for example, been applied to collision detection for autonomous vessels. There is also some research into mapping shipping behavior to detect anomalies or vessels that deviate from expected routes or shipping lanes. For example, this approach has been applied to detect illegal fishing activities.
The most relevant research regarding my thesis is work related to predicting vessel destinations in larger time intervals on a global scale. These approaches usually compare vessels' current trajectories to all historical trajectories from the same port or use port frequencies to construct a probability matrix of next destination ports from any given port.
There are some, but not many, existing research projects within this area. Some projects have been able to predict destination ports and regions to a reasonably high degree of accuracy. However, these related research projects seem to be limited in terms of data depth as they don't consider much detail, or features, of vessels such as dimensions, tonnages, draft, etc. This means that, for example, a dry bulk cargo vessel's data could be compared to historical data created by passenger vessels and vice versa. As different vessels travel in various patterns and ports, it would seem that these methods could be improved by increasing the data depth, such as making them aware of different vessel segments.
The approach will first consist of replicating an already established vessel destination prediction method derived from the existing literature. As mentioned above, the current work does not consider much detail about the vessels. Therefore, I will use Maritime Optima's novel segmentation of vessels, thus improving the existing method and achieving better predictions. Another goal of this work is to develop a foundation of a prediction model that can iteratively be improved upon by adding more data depth. Maritime Optima has an extensive amount of data for every vessel in their database that provides a unique opportunity to keep experimenting and improving vessel destination predictions even after completing the Master's thesis.
There will almost certainly be several challenges throughout the upcoming project, although I think the project is feasible given all of the available resources of both Maritime Optima and NTNU. The biggest concern will probably be related to the quality and extent of the data, as it is an unknown factor to some degree. The quality of the data will have an enormous impact on the prediction model's quality and performance. However, low data quality might be possible to use by finding a "work-around" and combining data from different sources. Therefore, an essential part of the project will consist of analyzing the data to determine the coverage and quality of the AIS data available.