Research On Ship Trajectory Prediction Method Based On AIS Data

With the increasing globalization of the world economy,maritime transportation for economic and commercial purposes is becoming more and more prevalent,which will bring certain pressure to maritime transportation,such as rapid increase in traffic volume,increased pressure on waterways,increased number of ships,and increased density of ships in critical waters and waterways.Under the above circumstances maritime accidents will occur frequently.To ensure the safety of maritime navigation and reduce traffic accidents,the key issue is to explore the ship navigation law and track reliability prediction.Therefore,accurate trajectory prediction is of great significance to reduce maritime traffic accidents and improve the decision-making level of related maritime departments and ship traffic service systems.Automatic Identification System(AIS)equipment is a new type of navigational aid system and equipment,which is installed on almost all ships.Compared with the traditional Vessel Traffic Service(VTS)data and radar data,AIS data are more easily obtained.In order to verify the performance of the proposed trajectory prediction algorithm,this paper uses the data-driven perspective and combines the neural network related algorithms to train the model and predict the time-series related ship trajectory sequences using marine AIS data and inland AIS data,respectively.The main contents of the thesis are as follows:(1)AIS data pre-processing.For the situation that the original AIS data is large and contains noise,the data are firstly cleaned and trajectories are extracted.Secondly,the feature correlation and trajectory smoothness analysis are carried out to select the input feature attributes suitable for the prediction stage and provide reference for the subsequent design of a reasonable prediction model.Since the time interval of the extracted trajectory point sequences is not continuous,this paper selects three times spline interpolation for trajectory repair to ensure the maximum restoration of the original trajectory data.(2)A one-dimensional convolutional network and bi-directional gated recurrent unit(1DCNN-Bi GRU)based ship trajectory prediction method is proposed.Long Short-Term Memory(LSTM)and Gated Recurrent Unit(GRU)are transferred to the ship trajectory prediction,which belongs to the same field of time series prediction,and the overall structure of 1DCNN-Bi GRU model is designed by invoking 1D-CNN.The predicted results are significantly better than the single LSTM trajectory prediction model by experimental validation.(3)A study of a generative adversarial network-based method for predicting ship trajectories is presented.Given the advantages of Generative Adversarial Network(GAN)which does not depend on data distribution,the adversarial learning method is used for predicting ship trajectories.A 1D-CNN network is chosen for the generator and a GRU structure is chosen for the discriminator.By adding the Gradient Penalty(GP)term,the overall structure of the GRU-WGAN-GP model can effectively predict ship trajectories and improve the generalization power of the model.