Research And Application Of Geographical Spatio-temporal Outlier Detection Method

Spatio-temporal outliers detection is one of the tasks of spatio-temporal data mining.Spatio-temporal outliers are the spatio-temporal entities that are significantly different from their spatial and / or temporal neighbourhood thematic attributes.Spatio-temporal outliers detection can find small parts of the data that deviate from the whole mechanism.The research of this part of the data can find some potential information and rules,which provide the basis for early warning and decision making.The spatio-temporal outliers detection method plays an important role in public health,environmental protection,traffic planning,criminology and disaster science.Spatio-temporal outlier is derived from the space outlier,which taking temporal dimension into consideration on the spatial outlier.Spatial outlier is proposed by Shekhar to extract the mutation structure in geospatial data.In this paper,the spatio-temporal outlier detection method is studied,the main research contents include:(1)Based on the basic principle of genetic algorithm,combining with the spatio-temporal scan statistics,the log likelihood ratio in the scan statistics is introduced as the fitness fuction,and the spatio-temporal scanning window is used as a chromosome,using population evolution which includes selection,cross,variation and elitism can find the spatio-temporal outliers.Different forms of spatio-temporal outlier are designed for different form of scanning window,which are mainly divided into cylinder,variable radius cylinder and titling cylinder,which can detect the three different shapes of spatio-temporal outlier detection.The simulation data set is used to test the detection ability of the method.At the same time,application and analysis of this method for hand-foot-mouth disease in Guangdong province,and compared with the classic Kulldorff's scan statistics method.(2)Spatio-temporal data is divided into a series of spatio-temporal subsequences though temporal windows.The fuzzy C-means method is used to identify the clusters in each time window,and the clusters in each time window are anomalously scored.In the fuzzy relationship,the abnormal dynamic evolution process will be visualized at successive time intervals.Among them,dynamic evolution can be divided into: emergence,disappearance,division and merger.At the same time,this method was applied to Shandong PM2.5 air quality monitoring data.The dissertation focuses on spatio-temporal outliers.The former focuses on spatio-temporal outliers detection of different shapes from the perspective of space-time.The latter focuses on the dynamic evolution process of spatio-temporal outliers.Through testing,the former is adapted to detect compact clusters of anomaly because of the shape of sanning windows,but has poor detection capabilities for irregular shapes and strips.The latter can detect the dynamic evolution of spatial anomalies.The relevant research results have certain theoretical and application values.