Research on a Berth Handling Type Recognition Model Based on Multi-Source Maritime Data

Existing studies on the recognition of berth handling types have a series of shortcomings such as narrow data coverage, low model precision, and inability to dynamically monitor global trade, the overcoming of which requires the development of more scientific and intelligent recognition methods. To solve these problems, this paper proposes a berth handling type recognition model. The first step is to build a model of geospatial correlation between the origins of strategic materials and berths, and recognize export berths based on the nearest distance and the Local Moran's I. Then a multi-view semi-supervised machine learning model is constructed based on the berth-vessel interaction network (BVIN) and the berth infrastructure sharing network (BISN), and used to infer the import berths on the list of berths associated with export berths. Modeling and experiments are performed in this paper using the berth handling type recognition model by taking soybean as a case study. The global vessel arrival and departure data, the arrival and departure data of soybean-carrying vessels related to China, and other relevant data of 2023 are collected. According to the findings of this paper, the model can identify 82% of soybean export berths, which are mainly distributed in major exporting countries such as Brazil, Argentina, and the US. The soybean import berth prediction model has a precision of 79.1% and an F1 score of 0.725, covering major importing countries such as China, Japan, and the EU. The above findings suggest that the algorithm can effectively recognize the main soybean import and export berths and clarify the relation between berths, providing a basis for the rational allocation and dynamic monitoring of port resources.