基于余弦相似度和移动曲面拟合的地面气温质量控制算法  

A quality control algorithm for cosine similarity and moving curved surface fitting

作　　者：叶小岭[1,2] 姚锦松 熊雄[1,2] 巩灿灿 金瞳宇 YE Xiaoling;YAO Jinsong;XIONG Xiong;GONG Cancan;JIN Tongyu(School of Automation,Nanjing University of Information Science & Technology, Nanjing 210044, China;Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters, Nanjing University of Information Science & Technology, Nanjing 210044, China)

机构地区：[1]南京信息工程大学自动化学院,南京210044 [2]南京信息工程大学气象灾害预报预警与评估协同创新中心,南京210044

出　　处：《中国科技论文》2020年第5期577-584,592,共9页China Sciencepaper

基　　金：国家自然科学基金资助项目(41675156);江苏省高等学校自然科学研究面上项目(19KJB170004);中国铁路总公司科技研究开发重点课题资助项目(N2019T003)。

摘　　要：针对传统地面气温质量控制算法在选取参考站和赋权时大多采用与目标站的欧式距离来判断与目标站的相关性程度,易导致最终质量控制结果精确度低且不稳定的问题,提出一种基于余弦相似度和移动曲面拟合的地面气温资料质量控制(MSF-COS-RMSE,MCR)算法。通过引入余弦相似度函数和均方根误差(root mean square error,RMSE)分析邻近站与目标站观测数据相关性,并使用余弦值与RMSE值对各参考站进行赋权,结合移动曲面拟合(moving surface fitting,MSF)算法拟合空间相关性曲面,求取目标站气温估计值。利用所提MCR算法对国家气象中心提供的8个地面观测站2014年日均温观测数据进行质量控制分析,并与反距离加权(inverse distance weighting,IDW)算法、空间相关性及B样条曲面拟合(spatial correlation and B-spline surface fitting,BSF)算法、MSF算法以及空间观测差异的地面气温资料质量控制(spatial observation diversity fitting,SDF)算法进行对比,实验结果表明:MCR算法的质量控制效果优于IDW算法、BSF算法、MSF算法及SDF算法,能有效地标记出气温观测数据中的可疑值,具有稳定性高、适应性强的特点。To solve the problem that traditional quality control algorithms of surface temperature easily leads to low precision and unstability in the selection of reference station and empowerment mostly using Euclidean distance to determine the degree of correlation with the target station,an algorithm based on cosine similarity and moving curved surface fitting of the ground temperature data quality control was proposed.Firstly,a cosine similarity and root mean square error(RMSE)analysis of adjacent station and target station observation data correlation was introduced,and the cosine and RMSE values was used to empower for the reference station.Secondly,the curved surface of spatial correlation was fitted and the estimate of target temperature was calculated combining with a moving surface fitting(MSF)algorithm.Finally,based on MCR algorithm,eight ground-based average daily temperature observation data of 2014 provided by National Meteorological Center was analyzed for quality control,and was compared with inverse distance weighting(IDW)algorithm,spatial correlation and B-spline surface fitting(BSF)algorithm,MSF algorithm,and space observation differences ground temperature data quality control algorithm(spatial observation diversity fitting,SDF).The experimental results show that the quality control effect of MCR algorithm is better than that of IDW algorithm,BSF algorithm,MSF algorithm and SDF algorithm,and it can effectively mark the suspicious value in the temperature observation data with high stability and strong adaptability.

关 键 词：余弦相似度 移动曲面拟合 均方根误差 地面气温 质量控制 

分 类 号：P412[天文地球—大气科学及气象学]