The growing demand for maritime transportation leads to increased traffic in ports. From this arises the need to observe the consequences of the specific speed ships reach when approaching seaports. However, usually the analyzed cases refer only to the statistical evaluation of the studied phenomenon or to the empirical modelling, ignoring the mutual influence of variables such as ship type, length or weather conditions. In this paper, a different approach to the issue is proposed, which summarizes the most important factors and leads to a probabilistic speed model for manoeuvring ships in the port of Gdańsk. For this purpose, data from the Automatic Identification System were used. This resulted in a dataset with almost 2.5k traffic scenarios. To obtain results from the dataset, three different machine learning algorithms based on Bayesian networks were then applied. The developed models can be used to predict the speed as a function of the given parameters as well as to determine the values of individual parameters for a given speed. In addition, the use of the constructed models allowed the analysis of the strength of mutual influences for two connected nodes or the sensitivity of changes for individual variables. The discussion also raised questions about validation of the algorithms and measures to improve accuracy. The average predictive accuracy of the models of about 75% (depending on the learning algorithm used) achieved at this stage is promising, but further work is expected that can increase the predictive power of the models.
Authors
Additional information
- Category
- Aktywność konferencyjna
- Type
- publikacja w wydawnictwie zbiorowym recenzowanym (także w materiałach konferencyjnych)
- Language
- angielski
- Publication year
- 2023
