Research On Data-based Trajectory Segment Clustering And Route Prediction Methods

With the increasing number of various types of ships,the state of water traffic in our waters is becoming more and more complex,which also increases the probability of collision of ships during navigation.By predicting the future trajectory of ships in the water,we can effectively avoid water collision of ships.In the past few years,the construction of AIS system in China has become more and more perfect,and more satellite and shore-based AIS receiving equipment has been put into use,which makes the data quality of AIS data guaranteed.This paper uses AIS(Automatic Ship Identification System)data as the research object,and mainly conducts in-depth research on the ship trajectory clustering method and ship trajectory prediction method based on data mining theory and deep learning theory.The main work is as follows.1)Based on the Hausdorff distance measurement theory,this paper reduces the impact of noise points on the distance measurement results in traditional methods by introducing One Class Support Vector Machine(OCSVM)and its internal similarity measurement method.Clustering is done using DBSCAN.In this study,the clustering results are evaluated using the proportion of effective trajectory segments and the silhouette coefficient index.The proposed similarity algorithm can not only improve the accuracy of the similarity measurement of the clustering algorithm,but also has obvious advantages in the measurement results of airline clustering.Subsequent trajectory prediction work provided high-quality datasets.2)This paper proposes a deep learning ship trajectory prediction method based on LSTM-MulAttention,which builds on an end-to-end prediction system that enables mapping learning between past and future trajectory sequences.This research method is based on the LSTM codec structure,and on the basis of the LSTM model,a multi-head attention mechanism(Multi-Head Attention)is innovatively introduced to optimize the trajectory prediction results.The verification results show that compared with the traditional LSTM model and LSTM-Attention model,the LSTM-MulAttention model proposed in this study has significantly improved performance such as trajectory prediction error.