隐含动态地理统计校准模型——以PM_(2.5)污染分析为例  

Hidden Dynamic Geostatistical Calibration Model——with Application to PM_(2.5) Pollution

作　　者：王雅琼 徐敏亚[1] 王菲菲 WANG Ya-qiong;XU Min-ya;WANG Fei-fei(Guanghua School of Management,Peking University,Beijing 100871,China;Center for Applied Statistics,Renmin University of China,Beijing 100872,China;School of Statistic,Renmin University of China,Beijing 100872,China)

机构地区：[1]北京大学光华管理学院,北京100871 [2]中国人民大学应用统计科学研究中心,北京100872 [3]中国人民大学统计学院,北京100872

出　　处：《数理统计与管理》2021年第2期191-204,共14页Journal of Applied Statistics and Management

基　　金：国家自然科学基金项目(72001205);中国人民大学科学研究基金(2021030047);国家重点研发计划《大气污染成因与控制技术研究》(2016YFC0207700,2016YFC0207701,2016YFC0207702);中国人民大学“统筹支持一流大学和一流学科建设引导专项资金”;全国统计科学研究项目(2019560);教育部人文社会科学重点研究基地重大项目(17JJD910001).

摘　　要：近年来,伴随着快速的工业化以及大量的能源消耗,PM_(2.5)污染成为突出环境问题,各地区雾霾现象频现.探索PM_(2.5)污染物的时空演变过程及形成机制受到越来越多的关注.与此同时,由于实际中电压波动、气候异常、仪器故障、不当维护等因素的影响,空气污染监控网络站点中的数据可能会偏离真实值.因此,本文提出了一种双水平分层时间空间模型,即隐含动态地理统计校准模型,并使用卡尔曼滤波器和期望—最大化(EM)算法进行模型估计.该模型通过将动态随机场作为随机隐变量,来刻画空气污染物的变化规律,并且通过引入加法和乘法校正系数,来检验监控站点数据的偏差.本文将该模型应用于河北省36个站点的PM_(2.5)小时数据,通过控制其它污染物和气象变量的影响,对PM_(2.5)污染的影响因素给出了合理的解释,并且基于校正系数发现了读数准确性存疑的站点.In recent years,as the consequence of rapid industrialization and alarmingly increasing energy consumption,PM_(2.5) pollution has become an outstanding environmental problem,and smog frequently occurs in various regions.It has attracted more and more attention to explore the spatial-temporal dynamic and formation mechanism of PM_(2.5) pollution.Meanwhile,due to voltage fluctuations,climate anomalies,instrument failures,improper maintenance,etc.,air pollution data collected from monitoring networks,however,could deviate from the true value.Hence,we propose a two-level hierarchical spatialtem poral model,i.e.,the hidden dynamic geostatistical calibration(HDGC)model,with parameters estimated through the Kalman filter and Expectation Maxim ization algorithm . On the one hand, the calibration model introduces dynamic random fields as a random hidden variable to characterize the dynamic mechanism of air pollution;on the other hand, it facilitates to detect and calibrate monitoring sites with biased readings by incorporating the additive and multiplicative calibration coefficients. The method is demonstrated with application to hourly PM_(2.5) data from 36 sites in Hebei province, China.By considering the influence of other pollutants and meteorological variables, a reasonable explanation is given for the influencing factors of PM_(2.5) pollution. Besides, based on the analysis of calibration coefficients, sites with skeptical readings are detected.

关 键 词：时空模型 校准系数 EM算法 PM_(2.5) 

分 类 号：C81[社会学—统计学] O212[理学—概率论与数理统计]