| In recent years,with the popularity of location information collection equipment,the data mining technology of moving target trajectory is gradually hot and has achieved extensive research results.However,in the field of navigation,due to the characteristics of sparse ship trajectory data,low data quality,strong randomness of ship position,and large space-time span,the traditional space-time trajectory data processing technology is difficult to be effectively applied and popularized.The data mining of ship moving trajectory has a very important application value for ship navigation safety and port ship scheduling.Therefore,this paper designs and implements a ship trajectory analysis system based on data mining of AIS(automatic identification system)data collected during ship navigation.First of all,this thesis designs a method to identify the ports where the ship is docked based on stop points detection and trajectory clustering.Which port the ship is docked is very important for ship trajectory analysis and port information statistics,but AIS data does not contain this information.There are large recognition errors to identify the port by the distance between ship and port or with the location point clustering method.This thesis identifies the port based on stop point detection.At the same time,to solve the omission and misjudgment of the port,the thesis uses stop points clustering,port boundary generation,trajectory clustering and other methods to further improve the accuracy of port identification.Experiments show that the recognition rate of this method is better than the two benchmark methods.Secondly,this thesis designs a ship destination port prediction method based on trajectory similarity.The existing prediction methods based on key points are difficult to achieve high accuracy in the data set with a low sampling rate,while the prediction methods based on grid partition have low performance,which is not suitable for the global prediction of destination ports and real-time prediction based on AIS data stream.The prediction method based on trajectory similarity proposed in this paper uses not only a single location point,but also trajectory similarity features,ship features and state features.The accuracy of the prediction method and the improvement of prediction results with trajectory similarity features are verified by constructing two machine learning models,random forest and LightGBM.This thesis also implements the ship trajectory analysis system based on AIS big data,designs the AIS big data storage,calculation and visualization scheme,completes three main modules of ship position tracking,historical trajectory analysis,and port information statistics,and carries out function test and performance test to verify the practicability of the system. |
