Analysis And Prediction Method Of Vessel Trajectory

In modern society,economic and production activities accumulate a large number of trajectory data with different structures.It is urgent to excavate valuable information from the vessel' trajectories.In the ocean transportation market,the sailing schedule and oil consumption are closely related to the income,and also closely concerned by the ship operators.Therefore,according to the vessel's trajectories and the information about vessels,the sailing schedule or oil consumption will be prejudged,which will bring great convenience to the operation of vessel.Therefore,more and more attention has been paid to the research on the analysis and prediction methods of vessel trajectory data.The processing object of traditional data mining algorithms e.g.clustering,classification is independent data,and each trajectory has continuity in time and space.The ship trajectory data mining are mostly based on point,ignoring the continuous in time and space as well as the classification and recognition in different motion patterns(e.g.different speed or direction),which can not reflect the dynamic changes of driving speed and heading of vessels.At the same time,most of the vessel trajectory prediction methods are mainly based on historical vessel trajectory data,ignoring the impact of the vessel's own attributes and the dynamic change of shipping market on the ship's future trajectory.In this paper,a clustering algorithm based on the characteristics of ship trajectory is designed,and the ship trajectories are classified into different categories according to the difference of routes and the difference of speed.For the prediction of ship trajectory,this paper finds out the factors that influence the type of ship trajectory,and establishes the prediction model of ship trajectory.When we get the properties of vessels,e.g.the captain,breadth,and shipping market data e.g.tariffs,oil prices,we can predict the future trajectory of vessel using prediction model.The research of this paper includes the following aspects:(1)According to the characteristics of ship trajectory,a ship trajectory clustering algorithm is proposed.Based on the characteristics of the trajectory information,the trajectory data is pre-processed.Then,in order to achieve the purpose of data compression and trajectory data feature extraction,we segments the trajectory data according to the speed change,so that the amount of data is greatly reduced.In order to calculate the similarity between trajectories,we use dynamic time warping algorithm to match the sub-trajectories,and the sum of the structured distances between the sub-trajectories is the distance between the two trajectories.Finally,according to the uneven distribution of trajectory data density,an adaptive density clustering algorithm is proposed to process the trajectory data.(2)According to the factors that influence the ship's trajectory,a prediction model of trajectory type is established.The model uses the vessel's attributes and data of shipping market's status as input,and output is the types of vessel's future trajectories,which can predict the types of vessel's trajectory.The type of vessel's trajectory is based on the result of the vessel trajectory clustering.For the limitation of logistic regression algorithm under multi-classification,an improved multiple SoftMax algorithm(IM-SM)is proposed.This algorithm provides a solution to the problem of imbalanced data and low degree of inter class recognition.For the problem of low degree of inter-class recognition,IM-SM algorithm adds the degree of inter-class recognition,so that the recognition degree between classes can be increased effectively in the training process of the model,which improves the prediction accuracy of the model.By modifying the loss function,the IM-SM algorithm takes into account the characteristics of the training data in the training process,which greatly improves the classification effect of the model.