回归模型在地震前兆观测缺数问题上的运用  

APPLICATION OF REGRESSION MODELS TO FILL MISSING DATA IN EARTHQUAKE PRECURSOR OBSERVATION

作　　者：李晓东[1] 贾鸿飞[2] 邓卫平[2] 

机构地区：[1]新疆维吾尔自治区地震局,新疆乌鲁木齐830011 [2]中国地震局地壳应力研究所,北京100085

出　　处：《内陆地震》2016年第4期358-364,共7页Inland Earthquake

基　　金：中央级公益性科研院所基本科研业务专项(ZDJ2014-04);国家自然科学基金(41104051);2015年度全国地震监测预报专项重点任务(86#)

摘　　要：以川03井水位观测值(整点值)为例,采用赤池信息量准则(AIC,Akaike Information Criterion),分别运用多元回归与偏最小二乘法对观测数据缺数进行补全计算。通过与实际观测值进行对比发现,多元回归与偏最小二乘法均可很大程度地还原观测数据。对于缺数天数较少时,采用多元回归预测模型具有明显优势,但考虑到解释变量的多重共线性问题,PLS对于缺数天数较多时,其稳定性要优于多元回归。They used the muhivariable linear regression model （MLR） and partial least squares model （PLS）, by Akaike information Criterion （AIC） to chose the best model, to fill the missing data in water level observation data （hourly data） in Chuan No. 03 Well. The results displayed that MLR and PLS all can restore the missing data in a large degree. The best way is MLR when only few days＇ data were missed, however, PLS was better than MLR when several days＇ data were missed. In additional, the PLS was close to practical process if considering muhicollinearity.

关 键 词：多元线性回归 偏最小二乘法法 缺数补全 地震前兆 

分 类 号：P315.7[天文地球—地震学]