| The rapid development and widespread application of the mobile Internet and Global Positioning System(GPS)have led to an explosive growth in Spatio-temporal data.A large number of Spatio-temporal data including the space,time and special properties of Spatio-temporal objects can provide a basis for the occurrence of major events.The analysis of Spatio-temporal data has become a hot topic in current research.The mining of abnormal Spatio-temporal data can directly analyze the occurrence of abnormal conditions.It is of practical significance to study the mining of abnormal data.Clustering analysis based on Spatio-temporal scanning statistic is one of the main methods of anomaly data mining.Many scholars have carried out research and obtained some results,but still have the following defects: First,the threshold setting of Spatio-temporal scanning window parameters will affect the scanning results,how to Setting a reasonable threshold is a problem that needs to be discussed.Second,the traditional space-time scanning statistic method does not consider the interaction of time and space,and the scanning result leads to low detection sensitivity.Therefore,the following research is done on these two problems:(1)In the traditional space-time scanning method,the spatial and spatial object space attributes are not considered.The spatial scanning step size and the scanning maximum radius threshold are fixed values during the whole spatial scanning process,which affects the accuracy of the scanning analysis.In this thesis,a space-time scan based on non-uniform step size is proposed.The maximum distance between the window level and other collection points is taken as the maximum spatial scanning radius of the window level.The spatial scanning adopts the progressive scanning method.It is proved that the detection analysis under this method is more reasonable and the result is more accurate.(2)This thesis proposes the C_Space-Time Scan algorithm.Firstly,in the time dimension,the algorithm uses the exponential weighted average method to superimpose the influence of the current time unit on the current stage to calculate the expected occurrence of the current time unit;then calculate the neighbor of the Spatiotemporal object based on the inverse distance weighting method in the spatial dimension.The effect of the domain on it,and finally the expected occurrence of the Spatio-temporal object in the current spatial position of the current time unit.(3)Based on the Python language,this paper completes the algorithm implementation,and uses the public data of the personal medical information of the patients with fever in New York City in the United States and the fever dataset of the border counties in Yunnan Province to conduct comparative verification analysis based on the definition of abnormal data.Verify the correctness of the abnormal point discrimination and the validity of the algorithm.This thesis improves and studies the traditional Spatio-temporal scanning statistic method from the perspective of scanning mode and space-time interaction.The results show that the improved spatio-temporal scanning method has higher detection capability,and the relevant analysis results provide guidance for practical applications. |
