The expansion of maritime transportation has resulted in increased traffic flow, complex routing, frequent encounters, and a higher risk of accidents in crowded waters. Accurately understanding the behavioral patterns of surrounding vessels has become essential for ensuring the safe and efficient of maritime traffic systems. The Automatic Identification System (AIS) provides abundant trajectory information—including latitude and longitude, course, and speed—which offers a large amount of data for ship behavior modeling and representation to support maritime situation understanding. Such research is of great significant importance for advancing the development of intelligent maritime transportation systems. Based on historical AIS data, this paper analyses the multidimensional behavioral characteristics of ships (position, course, and speed) and proposes a ship behavior modeling method based in semantic modeling techniques, then, a ship behavior representation method is developed using the long short-term memory-variational autoencoder (LSTM-VAE) model. Furthermore, kernel principal component analysis (KPCA) and cluster techniques are employed to extract ship behavior feature sequences. This study contributes to the theoretical and practical foundations of maritime traffic safety management and support the advancement of intelligent maritime transportation.