Ship Trajectory Prediction Model Based on Generative Adversarial Network with Attention and Interaction

Ship trajectory prediction is essential for finding the risk of collision and planning collision avoidance maneuvering. The ship trajectory prediction model based on the GAN-AI（Generative Adversarial Networks with Attention and Interaction） can make the prediction more accurate in multi ship encountering situations. The time-space sequence of the ship trajectory is encoded by an encoder and the relative positions and speeds of target ships are acquired and analyzed by the “interaction module”. The “attention module” integrates the motion information of own ship and target ships and feed the information to a resolver which predicts the ship trajectory. The predictor is verified with historical ship trajectory data from Zhoushan port. Test results show that this design is more accurate than Seq2 seq, simple GAN, and Kalman model by 20%, 24%, 72%,respectively.