Application of Bi-GRU for ship trajectory prediction
To improve the accuracy of ship trajectory prediction,this paper introduces a Bi-directional Gated Recurrent Unit(Bi-GRU)neural network,which is capable of mining sequence features in-depth,and constructs a ship trajectory prediction model based on Bi-GRU,starting from the shortcomings of Gated Recurrent Unit(GRU)neural network in extracting sequence features.The Automatic Identification System(AIS)data of 15 consecutive days in the waters near the Dashengguan Yangtze River Bridge in the Jiangsu section of the Yangtze River main channel were used as a case study to verify the accuracy and reliability of the Bi-GRU prediction model in predicting ship trajectories,and it was compared with Back Propagation(BP)neural network,Recurrent Neural Network(RNN),Long-Short Term Memory(LSTM)neural network,GRU neural network and Convolutional-Gated Recurrent Unit Neural Network(CNN-GRU).The results showed that the error margin of the BP neural network prediction results was too large and the distribution of points was too scattered to keep the error within an effective range.Although the overall errors of LSTM,GRU,and CNN-GRU were smaller,their error metrics were still larger than those of Bi-GRU.Bi-GRU could control the errors of most of the predictions between-0.001 5 and 0.001 5,and the three error metrics of Bi-GRU were smaller than those of other neural network models under the same training conditions.The Bi-GRU model performed best in this case when the learning rate was 0.005.Compared with BP neural network,RNN,LSTM,GRU and CNN-GRU,the Mean Square Error(MSE)of Bi-GRU was reduced by 80.4％,86.7％,9.4％,13.0％,and 49.4％respectively,the Root Mean Square Error(RMSE)was reduced by 55.7％,55.7％,4.4％,6.5％,and 28.7％respectively,the Mean Absolute Error(MAE)decreased by 70.4％,77.1％,7.2％,16.5％and 13.0％respectively.The error distribution of Bi-GRU is more concentrated,which proves the effectiveness and stability of the Bi-GRU ship trajectory prediction model.The research results of this paper can provide theoretical support for improving the safety management level of the ship traffic service system,judging the level of ship traffic risk and intelligent ship collision warning.
NETL
NSTL
万方数据
维普
