| The shipping industry is the foundation of the global logistics system,carrying out more than 80 percent of the world’s trade.Ship congestion caused by frequent maritime trade has brought new challenges to maritime traffic supervision and navigation safety.Therefore,accurate trajectory prediction plays a key role in Marine safety assurance.In recent years,the application and popularity of Automatic Identification System(AIS)makes it possible to accurately predict ship trajectory.Using AIS data to deeply mine trajectory information,modeling and predicting ship trajectory has become a hot topic in current research.However,current research focuses on the feature extraction and prediction of a single ship trajectory,and the model lacks universality.Aiming at the limitations of current ship trajectory feature extraction and prediction,this thesis proposed a general ship trajectory prediction model,which mainly includes AIS data processing,feature extraction and trajectory prediction.The main research contents are as follows.Firstly,in AIS data processing.For the original AIS data,this thesis processed the missing of field information and trajectory fragments,the accuracy of field and motion logic(such as latitude and longitude drift,course or speed drift),and the refinement of field and trajectory data.Secondly,in feature extraction.We improved the Density-Based Spatial Clustering of Applications with Noise(DBSCAN)algorithm and improved point clustering to trajectory clustering.The Dynamic Time Warping(DTW)algorithm was used to measure the similarity between different trajectories,the trajectories were divided into trajectory groups with obvious common characteristics,and then we performed group analysis of similar trajectories.Then,this thesis used the idea of frame segmentation to solve the problem of different number of trajectory points in the same trajectory group caused by different frequency of data acquisition.Finally,we improved the traditional single trajectory feature extraction algorithm propose group hierarchical clustering(GHC)for feature extraction based on trajectory groups,which can represent trajectories as sequence of the same number of feature points for data alignment and feature fusion.Thirdly,in trajectory prediction.We proposed a general prediction model of ship trajectory based on the attention mechanism.This model changes the input weight from the perspective of allocating input weight,which helps to improve the accuracy and generalization of prediction.In the experimental verification part.Firstly,in the data processing experiments.We visualized the data before and after processing.Secondly,in the feature extraction experiments.This thesis compared DBSCAN algorithm with traditional clustering algorithm,which proved the effectiveness of DBSCAN algorithm in the application of trajectory grouping.This thesis discussed and optimized the frame length to ensure the unity of the number of trajectory points in the same group of trajectory.This thesis compared GHC algorithm with traditional trajectory feature extraction algorithms(such as Douglas-Peucker algorithm and hierarchical clustering algorithm).Through quantitative comparison of evaluation indexes,we found that the trajectory recovery degree of GHC feature extraction algorithm was higher.Finally,in the trajectory prediction and application experiments.This thesis compared the current research hotspot models and the combination of hotspot models and different feature extraction algorithms.Through the comparison of evaluation indexes,it was found that the model proposed in this study had the best Mean Absolute Error(MAE).In addition,the MAE of the combination of feature extraction algorithm and attention mechanism model proposed in this thesis is optimal.And then this thesis completed the application test of the model,the results show that the model is effective in practical application.In summary,this research has laid the foundation for shipping digitization and shipping information service.It has important guiding significance and research value for maritime supervision,ship risk assessment,ship flow monitoring and other research fields. |
