Spatio-Temporal Change Point Detection Based On Self-Normalization Method

With the continuous acceleration of urbanization and industrialization,China is facing serious environmental pollution problems,especially air pollution,which has received increasing public attention in recent years.China determines air pollution mainly through data from pollutant monitoring stations around the country,so the accuracy of monitoring station data is particularly important.We hope that the monitoring data abnormalities in time to find and troubleshoot the monitoring instrument,but when the pollutant monitoring data abnormalities may be caused by wind,rain and other meteorological factors,so when the pollutant monitoring data abnormalities,it is not necessarily due to the failure of the monitoring instrument,if it is caused by meteorological factors is not necessary to investigate,so we need to distinguish when the monitoring station data abnormalities.Therefore,we need to distinguish whether the abnormal data at monitoring stations are caused by meteorological factors or human factors,and if they are human factors,they need to be corrected in time to ensure the accuracy of the data at monitoring stations.At present,the widely used verification method is manual verification of all monitoring stations with abnormal data by special employees,which is not only inefficient but also with obvious subjectivity.In order to improve the efficiency,this paper proposes a new spatio-temporal change point detection method based on the spatio-temporal characteristics of pollutant monitoring data using the Self-Normalization method.Firstly,the change point detection of spatio-temporal series data is performed based on the Self-Normalized variation point detection method,and then the spatial correlation between the target site and the neighboring site data is used to construct the test statistic,and finally the Block Bootstrap method is used to perform hypothesis testing to determine the type of site variation points.The method can effectively distinguish between normal change points caused by meteorological factors and abnormal change points caused by human factors,and the method uses the Self-Normalization method for change point detection,which avoids the situation that CUSUM affects the variation point detection results due to the existence of non-monotonic power when performing change point detection.In this paper,according to the different types of spatio-temporal data change points,the algorithm for detecting the mean change point and variance change point of spatio-temporal data is constructed based on the Self-Normalization method,and the validity of the algorithm is tested using simulated data,and compared with the spatio-temporal variation point detection algorithm constructed based on the CUSUM method,and the results are found:For the mean change point,the spatio-temporal variation point detection algorithm based on the CUSUM method exhibits the nonmonotonic power,while the spatio-temporal change point detection algorithm based on the Self-Normalization method does not exhibit the nonmonotonic power and has a high accuracy rate to effectively test the variation points.For the variance change points,both the spatio-temporal change point detection algorithm based on the CUSUM method and the spatio-temporal change point detection algorithm based on the Self-Normalization method have high testing accuracy,and both can effectively test the variation points.By modeling the actual PM_2.5data in Chengdu,we also find that the spatio-temporal data anomaly change point detection algorithm proposed in this paper can accurately identify the normal and abnormal change points in the actual data,which provides a certain basis for practical application.