Research On Key Techniques Of Anomaly Detection In Sea-air Target Trajectory Based On Big Data

As various types of sensors heavily installed in mobile terminals,spatio-temporal data accumulates a lot,and how to find information from massive data effectively is a hot topic of current research.As an important branch of spatio-temporal data mining,anomaly detection has important applications in both the military and civilian fields including financial fraud,climate prediction,animal migration,taxi abnormal driving detection,etc.especially,it can help the decision-makers to respond to the situation in a timely and effective manner by anomaly detection to grasp the initiative of the battlefield.At present,anomaly detection is mainly concentrated in the road network area with time and space constraints,such as taxi anomaly detection,and the research on anomaly detection of target in unconstrained free space is poor,especially the large-scale and unconstrained characteristic of sea and air targets in high-degree-of-freedom space make anomaly detection difficult.Based on the trajectory data of sea and air targets,this paper introduces the related methods of spatio-temporal data mining and studies the real-time trajectory anomaly detection technology suitable for the characteristics of spatio-temporal data of sea and air targets,and proposes the evaluation criteria of anomaly detection algorithms,which is convenient for comparing the performance of different models.The basic knowledge of trajectory anomaly detection is introduced.The principle and definition of anomaly are given.Based on the spatio-temporal data characteristics,the preprocessing method for spatio-temporal data of sea-air target is introduced.The model evaluation criterion of anomaly detection algorithm is proposed to realize the quantitative evaluation of anomaly detection model.In the daily monitoring,the false alarm rate for the anomaly detection of sea and air target data is too high.Based on this,a region-modeling based trajectory anomaly detection algorithm is proposed.It focuses on the overall movement trend of the target rather than the details.First,the target trajectory is transformed into a regional sequence;then the sequential pattern mining method is applied to model the transfer relationship between regions;next,based on the actual data and model,the target’s the next active region is predicted;finally,the comparison between the predicted result and the actual result is make to determine whether the trajectory is abnormal..In the tension situation,Based on the demand of the anomaly detection algorithm with high recall rate,a segment-based trajectory anomaly detection algorithm is proposed.Having get the abnormality of the target motion trend,the algorithm finds the details of the target anomaly.First,the trajectory is divided into trajectory segments,and the similarity metric matrix of the trajectory segment is constructed;then,get the abnormal trajectory segment by calculating the number of adjacent tracks,and further the abnormal trajectory is found;finally,the algorithm is modified to apply to the real-time anomaly detection of the trajectory data stream.Build and improve the spatio-temporal data mining system.The system is improved based on the original system,including basic modules such as data management,data situation analysis and data preprocessing,main modules such as frequent trajectory mining,group rules mining,trajectory prediction,and anomaly detection,as well as modules of the visualization for human-computer interaction.The module implements the complete process from raw data to knowledge output.The experiment results proves that the system respectively satisfies the special needs of abnormal detection in daily monitoring and tension situation,and can provide reliable criteria for decision makers in a timely and effective manner.