Data mining approach for automatic ship-route design for coastal seas using AIS trajectory clustering analysis

Abstract

In this paper, we propose an automatic route design method based on simple recurrent unit (SRU) and automatic identification system (AIS) data. Laplacian eigen maps and Gaussian kernel functions are used to compress the AIS data and extract the turning points of all ships. Fuzzy adaptive density-based spatial clustering of applications with noise (FA-DBSCAN) technique is used to cluster the turning points obtained at the preprocessing stage to obtain the turning region. Optimal turn region matching is used to connect the turning regions of similar routes, and the SRU neural network algorithm is used to learn the relationship between different types, sizes, and drafts of ships in each turning region; extract the feature-turning points; and obtain the recommended coastal routes, speed, and course of each type of ship. In the experimental stage, a large variety of AIS data from two sea areas are used to compare and analyze the designed route and real-ship data through LSTM and SRU experiments. The results show that the SRU algorithm improves the training speed and accuracy in comparison to LSTM, while the generated automatic route meets the requirements of navigation practice.