A Research On Trajectory Data Cleaning Based On Temporal Feature

With the advancement of sensor,network and positioning technology,all kinds of positioning equipment are widely adopted,generating a large amount of trajectory data.These data could reflect the activity characteristics of moving objects,and have great research and practical value in various fields,such as urban planning,business decision,traffic monitoring,etc.However,the steps carried out during the using of trajectory data,including collecting,transmission,storage and processing,may sometimes go wrong and introduce noise to the data,then interfere with subsequent research and application.Therefore,it is necessary to perform noise cleaning in advance.Existing trajectory data cleaning methods fall short in treating abnormal trajectories and cooperating with domain knowledges.On the one hand,abnormal trajectories are often detected and repaired as noise,which has a misleading effect on subsequent processing,research and application of the abnormal data.How to distinguish noise from abnormal trajectories is a difficult research problem.On the other hand,it is difficult for domain experts to apply domain knowledge to the cleaning process,and to adapt cleaning methods to different trajectory data sets based on the domain knowledge.To solve these problems,this paper explored data cleaning theories and methods that keep abnormal trajectories in the data sets,and put interpretability and domain knowledge in equal positions.This paper has three contributions:(1)A new trajectory data temporal feature extraction algorithm is proposed.Based on the local features of the time series of a trajectory data,this algorithm extracts fragments from the trajectory,and provides a series of more interpretable data,which is convenient for domain experts to understand,describe,and apply domain knowledge to.(2)A noise cleaning model that keeps abnormal data is designed.Based on the temporal features extracted from a trajectory data set,and the logical relations between the local and global data,this model makes a judgement for each trajectory on whether it is noise or anomaly,and offers suggestions on noise restoration.(3)The model and the algorithm proposed are highly practical,since they are data-driven and independent from any additional standardized information(such as road networks).(4)Weight parameters in the model are strongly interpretable,which can assist domain experts to clean different data sets according to their domain knowledge.Experiments are carried out,showing the effectiveness of the algorithm and the model proposed.