Ship Track Prediction Based On AIS Data And Improved LSTM Network

With the globalization of the world economy,the maritime shipping traffic flow has gradually increased.Accurate and reliable ship trajectory prediction can effectively guarantee the ship’s navigation safety.It is of great significance for improving the level of ship traffic management.Due to the ship’s own motion characteristics and external environmental factors,the ship’s trajectory has four characteristics of nonlinearity,randomness,trend and periodicity,which makes the ship’s trajectory prediction more difficult.The current ship trajectory prediction still has the following limitations: First,traditional mathematical modeling is difficult and the prediction error is large.Second,the large amount of trajectory data and high dimensions make feature extraction very difficult.Third,there are many factors affecting the ship trajectory.As a result,most models only discuss the effect of the ship’s own motion on the trajectory.Inspired by the limitations aforementioned,a navigation model matching the actual scene is trained to accurately predict the actual position of the ship in the next half minute or one minute in a complex environment.A short-term trajectory prediction model based on the seq2seq-Bi-LSTM network based on the ship trajectory data sets of Chengshan Cape AIS base station is proposed in this work.And verifying the effectiveness and accuracy of the method.The main ideas are as follows:First,after the comparison and analysis,the BP neural network model has better performance on the trajectory prediction rather than the other mainstream traditional trajectory prediction models,including multiple linear regression,Kalman filter,and gray prediction model.Second,through the pre-processing and visual analysis of AIS data,the characteristics of ship trajectory data are obtained.And the LSTM model is used to compare the influence of single input value and multi-input value on the short-term prediction model of ship’s trajectory.Then it is verified that LSTM has better performance of prediction than the BP neural network,and the model using multi-input value increases the adaptability and stability of the prediction model without losing the accuracy of the prediction.Third,it discussed the impact of the number of iterations on the prediction results adapting the short-term prediction model of ship trajectory based on LSTM based on three different types of ship trajectory data.From the experiments,it can be seen that as the number of iterations increases,the root mean square error decreases exponentially,and the accuracy of predication increases exponentially,the rate of accuracy increase is not obvious until the iteration times exceeded a certain threshold.At the same time,it is deduced that the volume of the ship is inversely proportional to the accuracy of the prediction from the results.Last but not least,Considering a two-way data exchange mechanism,an improved ship trajectory prediction model named the seq2seq-Bi-LSTM network is proposed.The module of Seq2 Seq can transform the AIS data into data sequences that can be processed by LSTM.And the main framework and details are given.In the end,the comparison experiments with BP and LSTM neural network models demonstrate the performance of predication from the metrics of maximum value,mean square error and computational efficiency.