This repo provides the material to support the work undertaken by the Data Science Campus to explore the operation, use and relationships between ports in the UK at a macro level and the behaviour and operational characteristics of ships at a micro level. Specifically we explore:
- ship travelling behaviours
- traffic at ports and related factors
- port capacity utilisation
- national and international port relationships
- inbound ship delays
The Automatic Identification System (AIS) and Consolidated European Reporting System (CERS) data sets, provided by Maritime and Coastguard Agency (MCA), were used for this study.
The AIS data records the position, speed, heading, bearing, rate of turn and other important information for each ship, at frequent time intervals throughout its voyage. The messages are received in raw encoded format using the NMEA standard. Decoding functions based on the description here, the AIVDM/AIVDO protocol decoding and the Navigation Center of Excellence message explanation were developed in Scala.
The CERS data set comes from an information management system maintained by MCA, that is intended to capture all UK ship arrival and departure notifications, dangerous and polluting goods notifications, and notifications of port waste infringements and bulk carrier infringements. This information then is forwarded to the EU SafeSeaNet system in accordance with the EU Vessel Traffic Monitoring and Information System Directive (2002/59/EC).
The study is implemented through a series of files that can be found in the ./code
folder. Make sure you read the ./code/config_explanation.md
first on how you should set up your config file, collect required data sets and name all your files. Then proceed as follow:
-
If your AIS messages are in a raw format then use the decoders in the
./code/1_AIS_decoder
folder to extract them in csv form. You can also find instructions to filter and process the decoded data inAIS_decoder.md
. -
If you have decoded AIS messages then jump to the
./code/2_preprocessing
folder where the AIS and CERS data sets are being formatted and the Rate of Turn (RoT) of a moving ship is calculated. -
The travelling behaviour of ships approaching port is explored in
./code/3_segmentation
folder. K-means unsupervised machine learning algorithm is used and six segments representing specific patterns are derived and explained. -
In
./code/4_feature_engineering
, we bring in new features for the delays modelling that will follow. -
Then we perform supervised machine learning using random forests, adaBoost and XGBoost to predict the probability of a ship being delayed in
./code/5_modelling
. We also uncover insights by exploring the relationship between the delay rate and each feature on a univariate basis. -
Finally we provide methods for different ports and voyage insights in
./code/5_modelling
folder. We look into port traffic, amount of delayed trips, movements between ports and movement of hazardous materials. Finally, we present a network analysis between UK and major international ports.
The complete report supporting this repo can be found here.