Feature Fusion And Abnormal Detection Of The Trajectory Data

In the intelligent and information society,satellite navigation and mobile phone positioning systems have developed rapidly.Atmospheric monitoring,smart commuting and other industries have collected massive trajectory data,and the stock of trajectory data will continue to grow.Trajectory data mining research is helpful to discover valuable hidden information in massive trajectory data.This paper focuses on the current status of trajectory data similarity measurement and abnormal trajectory detection algorithms.The main research contents are as follows:(1)The traditional trajectory similarity is only measured from the position vector,ignoring the speed and time characteristics of the trajectory data,which results in the trajectory measurement results can not effectively and comprehensively reflect the actual situation,reducing the effectiveness of the trajectory detection results.In order to overcome the above limitations,a trajectory data similarity measurement strategy for multiple feature vectors and a multi-feature fusion similar trajectory measurement method are proposed.The fused feature vectors are used to discover hidden important trajectory information.This method first maps the trajectory data to the trajectory graph described by the graph model.Each trajectory is a node of the trajectory graph.For the speed,time and space characteristics of each node,three kernel functions suitable for its metric are given.The weighted sum is used to realize the fusion of the three feature vectors;then the value of the feature fusion of each node is used to construct the similarity matrix of the trajectory data,so as to realize the similarity measurement of the trajectory data fusion features.Finally,the hierarchical clustering method is used to evaluate the feature fusion strategy of the trajectory.Experiments show that the strategy makes the clustering result more accurate,that is,the data in the same trajectory class has the same feature performance,while the data from different trajectory classes There are differences in feature performance.(2)We propose an anomaly trajectory detection algorithm TADSS based on sparse subgraphs,abstract the trajectory data into a graph model,and detect anomaly trajectory data by dividing the graph model and searching for sparse subgraphs.The TADSS algorithm first calculates the similarity value of the feature fusion metric between trajectory data to establish a trajectory vector feature map;then divides the trajectory feature map into multiple trajectory subgraphs by spectral clustering technology;finally,a novel concept of weight coefficient is proposed And define the sparse subgraph,search the sparse subgraph to realize the detection of abnormal trajectory,where the sparse subgraph is a special graph whose density is much lower than the average density value,so the data in the sparse subgraph can be regarded as an abnormal trajectory data.(3)We evaluate and verify the performance of the algorithm using multiple real data sets.First,using the trajectory data of taxis in Shanghai,the performance of similarity measures of trajectory data fusion was verified by clustering algorithm evaluation.The experimental results show that the clustering results after multi-feature fusion measures are more accurate and have good stability.Then use the vehicle trajectory data of Shanghai to evaluate the efficiency of the abnormal trajectory detection algorithm TADSS,and use the Atlantic hurricane data to evaluate the influence of the parameters on the experimental results.The experimental results show that the TADSS algorithm can accurately detect the real abnormal trajectory.Efficiency is better than existing algorithms...