Research On Ship Track Prediction Algorithm Based On AIS Information

In today’s economic globalization,the volume of international trade is growing,and most of these goods are transported by sea.There are more and more ships on the sea,which brings trouble to the ship’s driving.The automatic identification system(AIS)can automatically identify the surrounding ships and their navigation information,so that the ship can know the location and speed of the surrounding ships in advance.The ship track prediction based on ship automatic identification system can automatically predict the ship’s track in the next period of time,which is conducive to ship avoidance and plays an auxiliary role for the pilot.The neural network method can predict the ship’s track with high accuracy.As an important auxiliary system of automatic driving and manual driving,track prediction technology can be divided into long-term prediction and short-term prediction.Although this technology has been studied for many years,there is still room for improvement.Because the traditional genetic algorithm only improves the weight and threshold of the neural network,or only improves the number of nodes of the neural network,there is a problem that the number of nodes and the weight and threshold of the neural network can not be considered comprehensively,and the prediction accuracy is low.Therefore,in this paper,the short-term ship track prediction based on neural network is deeply studied,and a novel neural network training scheme of "genetic algorithm simultaneously improving the weight,threshold and node number of neural network + online learning" is proposed,and its feasibility is verified by experiments.The specific research of this paper is as follows:(1)Aiming at the problem that the traditional genetic algorithm only considers the weights and thresholds of neural network,this paper proposes a genetic algorithm method to improve the weights,thresholds and the number of hidden layer nodes of BP neural network.In this method,the number of nodes,weight and threshold of neural network are considered simultaneously,and a special coding method is proposed.The coding method overcomes the problem that chromosome length cannot be crossed when the number of nodes,weight and threshold are optimized at the same time.The experimental results show that the method has higher accuracy,lower mean square error,and the predicted track is closer to the real track.(2)In view of the problem that the off-line learning of BP neural network can not adapt to the change of ship navigation environment,and the incremental online learning has lower precision than batch learning in small training set,this paper proposes a batch online learning method.When the time limit of online learning is reached,batch online learning is used to train neural network,which solves the problem that offline learning can not adapt to the change of ship navigation environment.The experimental results show that the mean square error of the predicted track is reduced and the accuracy of the predicted track is improved.(3)In order to further apply the ship track prediction to practice,this paper designs the conversion program from Excel to KML file.The content and format of the transformed KML file are correct,the visualization function is realized,and the preliminary transformation of research results is completed.