GNSS Outliers Mining Algorithm Research

With the development of economy,trade and technology,GNSS is playing an increasingly important role in providing basic navigation and positioning services for global land,sea and air transportation.However,in the use of global satellite navigation system,the receiver usually gets abnormal positioning information for some reasons,which makes the users need to mine and judge the coordinate information outputted by the receiver in the subsequent application process,and get more reliable positioning data after removing or correcting the abnormal information.This paper aims at accurately identifying the abnormal positioning sequence in the driving track collected by GNSS devices,so as to meet the requirements of recognizing the driving state and predicting the driving intention in the future.The existing work classified the sequences by a single feature pattern,i.e.,only the segments which are different from the normal motion state are identified as the outlier sequences.However,the land traffic environment is more complex with diverse types of abnormal location sequences.Thus,the previous methods are not suitable in the abnormal sequence recognition of land traffic trajectory data.For different types of anomaly location sequences,different algorithms need to be designed according to their unique feature patterns.This paper focuses on three typical types of abnormal sequence and designs algorithms from the perspective of abnormal sequence type production scenarios.It overcomes the problem that the mainstream framework has difficulty to identify the abnormal location sequences in land traffic trajectory data to some extent.We use GNSS devices to collect more than 2000 kilometers of land traffic trajectory and analyze it.We roughly divided the abnormal sequences into three types:(1)anomaly positioning caused by multiple changes of direction at extremely slow speed;(2)anomaly tracking caused by location deviation in approximate uniform linear motion;(3)anomaly positioning caused by location deviation when the speed changes.Aiming at the above three situations,and combining with the generation scenarios and application requirements,we design the Dynamic-GMM algorithm based on dynamic curve fitting,the Consistency-K-Means algorithm based on motion consistency,and the Extremum-SVM algorithm based on feature extremum.Finally,the Dynamic-GMM algorithm is evaluated by V-Measure and Adjusted Rand Index,and the other two algorithms are evaluated by precision,recall and F1 score.The results clearly demonstrate the effectiveness of the proposed algorithms.In addition,we analyze the causes that led to the errors for improving in the future work.