Ship Track Prediction Based On Optimized Neural Network Model

With the frequent exchanges of international business and trade,the development of shipping industry has been promoted,which makes more and more ships join in the water transportation,and makes the traffic in some channels,harbors basins and other traffic intensive waterways busier.The increase in the density of ships reduces the distance between ships and increases the possibility of accidents such as collision and grounding.In order to reduce the incidence of maritime accidents and improve navigation safety,not only the maritime authorities need to further strengthen the supervision and guidance of vessel navigation,but also each vessel needs to know the navigation trend of the ship and other ships through all kinds of aids to navigation equipment,so as to carry out reasonable ship maneuvering.An important prerequisite for all of this is to predict the ship's future movement trend for a period of time,including track,speed,course and other dynamic navigation elements of the ship,among which the track information is the most important,so track estimation and track prediction methods have emerged to predict the position of the ship.In the trajectory estimation method,based on the mathematical model of the ship's movement,the establishment of the model requires more.ships and external environment parameters,and the ship parameters will vary with ship type and loading,and also the external environment will change from time to time.Therefore,the model does not have good generalization,and is difficult to be applied to other ships.Even if it can be used,it faces many problems such as complicated calculation and can only be used for the ship's straight path estimation.Therefore,the operation of track estimation is too complicated and limited in navigation practice.The application of neural network solves this problem well,making the trajectory prediction can predict the ship's trajectory without establishing a relevant mathematical model,and it can be applied to non-linear prediction and other advantages have been favored by many experts and scholars.Trajectory prediction uses the collected AIS time series data to import the established neural network prediction model,and fits the future trajectory information through the learning of the model.This method is convenient,fast,and highly accurate.But at the same time,many studies have found that there are differences in the learning ability of various neural network models.Therefore,the prediction accuracy will vary greatly with the different models.In order to learn the laws in the navigation data more accurately,improve the accuracy of the trajectory prediction,and ensure the safety of ship navigation.Aiming at the characteristics of time series and nonlinearity of navigation data,a hybrid model short-term trajectory prediction method based on convolutional neural network(CNN)and long-short-term memory network(LSTM)was proposed.This method combines the advantages of CNN and LSTM networks,takes a large amount of historical navigation information as input,firstly extract data features through a convolutional neural network,and then use the data features as input data of the long-term and short-term memory network in a time-series manner,and then predicts the ship track.Because the CNN-LSTM model uses the CNN model to extract the latent features of time series data,it provides effective input data for the LSTM model,thereby improving the accuracy of prediction.The method is used to make prediction experiments on real ship navigation data,and the experimental comparison of other models such as adaptive particle swarm optimization BP neural network(SAPSO-BP),LSTM,etc.The results show that the CNN-LSTM prediction model has higher prediction accuracy than other prediction methods;by analyzing the influence of different input number of continuous time,CNN convolutional layers and LSTM hidden layers on the result of prediction,and then obtain the best prediction model,then the model was tested based on complete track and compressed track,and better prediction results were obtained,indicating that the CNN-LSTM model has better prediction stability.