Research On Vessel Track Prediction Based On CNN-GRU

The extensive application of automatic identification system(AIS)around the world has made the analysis of ship navigation behavior through a large amount of historical and realtime AIS data a recent research hotspot.In order to improve the accuracy of ship track prediction and enable it to better serve maritime supervision,this thesis proposes a combined deep learning network based on AIS ship navigation data training,based on a twodimensional convolutional neural network(CNN)A combined neural network model was constructed with a gate recurrent unit(GRU)to predict the ship’s track.The main content of this article is as follows:(1)Based on the thermal map of water shipping,more than 40,000 AIS data were collected in the waters near Kukha Town in the Strait of Malacca,which were distinguished according to the tonnage of ships,and abnormal points were eliminated,missing data were completed,and three samples were used to complete the data.Bar interpolation performs the initial processing of equal time intervals on the data set,and obtains two data sets with equal time intervals of 20 seconds and 60 seconds.(2)Considering that CNN can extract the short-term feature relationship of the data set,and as a variant of the long-term and short-term neural network,the GRU neural network performs well in extracting long-term features,and can be combined with the CNN model to solve the problem of neural network models processing long-term The problem of gradient disappearance occurs in sequence data.This thesis estaablishes a CNN-GRU combined model ship track prediction model that considers five features including ship speed,course,and ship tonnage.(3)In order to verify the feasibility and accuracy of the model,the model was trained using the two processed data sets,and the convolutional neural network,long short-term memory neural network(Long Short-term memory,LSTM),gate Controlled cyclic unit neural network,combined neural network based on convolutional neural network and long-term short-term memory neural network,combined neural network based on convolutional neural network and gated cyclic unit,using MSE(mean square error)as the The evaluation index,at the same time comparing the predicted trajectory and the real trajectory of the 20-second interval data and the 60-second interval data,analyzed the accuracy of the trajectory prediction model.The longitude and latitude,course and speed in the ship AIS data are used as input data,and the longitude and latitude are used as output data to build a prediction model.The results show that the MSE of the test set of the CNN-GRU prediction model under the training of the two data sets reached 4.1368e-05 and 1.0796e-04 respectively,which is lower than the other three models mentioned above,and is more stable than the other three models,Accurately predict the ship’s track.Although the trained model cannot be directly used in other waters,the CNN-GRU model takes into account the speed and accuracy of the operation,can quickly fit the ship’s track,has strong robustness and portability,and can effectively assist the maritime system in obtaining information on ship traffic conditions.