Improved SocialVAE: A Socially-Aware Ship Trajectory Prediction Method for Port Operations

In comparison to other regions, the traffic dynamics observed in port waters are more unclear. Given the complexity of maneuvering in multi-vessel encounter scenarios, accurate prediction of vessel trajectories in port waters is crucial for preventing accidents and improving port operations. To address the highly dynamic navigational environment and the uncertainty of navigational patterns such as heading and path selection, we propose an improved SocialV Ae(Variational Autoencoder) method. The approach use local ship observations to comprehend traffic patterns between ships in order to make ship navigation decisions. Furthermore, we present a socially aware regression module that evaluates the compatibility between past trajectory interaction codes and forecast trajectories, and then uses the resulting offsets to adjust the predicted trajectories. Additionally, for the problem of non-random missing data in port waters AIS (Automatic Identification System) data, we use a shift-corrected data interpolation method to fill in the missing values. We utilize two real AIS datasets of waterways surrounding ports to test the efficacy of our method: Copenhagen Port and Shanghai Yangshan Port. When compared to existing methods, our Improved SocialV Aedelivers excellent prediction performance on these two datasets.