Research And Application Of Trajectory Similarity Relationship Mining Technology Based On Deep Learning

With the advancement and development of technologies such as mobile Internet and location information acquisition processing,many devices with positioning equipment(such as vehicles,ships,etc.)generate a large amount of moving target data.At present,the trajectory similarity mining research of spatio-temporal data has become one of the hot research directions of data mining.It has very important significance for users' personalized route recommendation,road traffic optimization,and climate change law analysis.Trajectory similarity relationship mining refers to calculating the matching degree between trajectories through different trajectory matching measurement methods.It is one of the most critical methods in the research of a series of problems such as trajectory clustering,trajectory anomaly detection,and user-trajectory link relationship mining.By tracing the similarity relationship of trajectories,the relationship between trajectories in spatio-temporal data can be vividly described,which provides a better reference for grasping the characteristics of the analysis objects.This paper makes full use of the technical advantages of current deep learning technology in semantic feature extraction and analysis,focusing on the study of trajectory similarity relationship mining,and its application and optimization in user-trajectory link relationship.The innovation of this article is mainly reflected in:1.This paper proposes a method of trajectory semantic information representation based on deep learning.The semantic information includes sequence dependency,time law,position preference and other characteristics in the trajectory.Point of Interest)position sequence information is input,and the POI position sequence is characterized by a vector based on the LSTM(Long Short-Term Memory)method,and the represented vector is used to determine the similarity between the trajectory and the trajectory.At the same time,the TULSN(Siamese Network for Trajectory-user Linking)model is used to solve the problem of difficult to accurately carry out trajectory and user linking due to privacy reasons for large-scale spatio-temporal data.Two parts,training and classification.Through training,the embedded representation of POI position sequence information generated by the same user is close,and the embedded representation of different users' POI position sequence information is far away;the KNN(k-Nearest Neighbor)method is used to find the most similar trajectory to the known user trajectory The user of the track.The paper conducts experiments on two public data sets to evaluate the performance of the TULSN model.The experimental results show that the TULSN model improves the user accuracy rate of identifying trajectories by 18.72% compared with the existing trajectory-user links.2.This paper proposes a method based on self-attention to obtain the most important POI points that can be used to distinguish trajectory targets for the situation where the scale of spatiotemporal data is large and it is difficult to find similar trajectories efficiently and different spatiotemporal points have different effects on the similarity relationship between trajectories.And give it a higher weight to improve the accuracy of trajectory semantic similarity calculation;at the same time,in order to improve the efficiency of similar trajectory retrieval in large-scale spatio-temporal data,the method of deephash is introduced,and the trajectory embedded representation based on deep learning is carried out to simplify and improve search efficiency.Experimental results show that the improved TULSN method is superior to the pre-improved TULSN method in accuracy and efficiency.