基于三种机器学习算法的降水现象仪和雨量筒数据一致性检验  

Data consistency test of precipitation phenomenometer and rain gauge based on three machine learning algorithms

作　　者：成振华 周坤论 陶伟[1] 黄剑钊[1] 王玮 景坤 Cheng Zhenhua;Zhou Kunlun;Tao Wei;Huang Jianzhao;Wang Wei;Jing Kun(Guangxi Meteorological Technology Equipment Center,Nanning 530022,China)

机构地区：[1]广西壮族自治区气象技术装备中心,南宁530022

出　　处：《气象研究与应用》2022年第3期115-119,共5页Journal of Meteorological Research and Application

基　　金：降水天气现象仪探测数据订正模型本地化研究(桂气科2022QN09)。

摘　　要：基于三种机器学习算法,对2018年南宁国家气象观测站雨量筒观测数据和降水现象仪的雨滴观测数据进行一致性检验试验。通过降维算法,对降水现象仪数据去除数据冗余,进一步分别采用多元线性回归、决策树回归、最近邻回归等3种机器学习算法验证与雨量筒数据的一致性情况。结果表明,综合性能中多元线性回归算法效果最好,在误差范围内的准确率达到85%以上;最近邻回归算法在小雨量中可以有较好的预测值,综合准确率达到75%,两种算法均优于决策树算法70%的准确率。Based on three machine learning algorithms, a consistency test was conducted on the rain gauge observation data and the raindrop observation data of the precipitation phenomenon instrument of Nanning National Meteorological Observatory in 2018. Firstly, the dimensionality reduction algorithm is used to remove the data redundancy of the precipitation phenomenon meter data. Three machine learning algorithms including multiple linear regression, decision tree regression, and nearest neighbor regression are further used to verify the consistency with the rain gauge data. The results shows that the multiple linear regression algorithm has the best effect in the comprehensive performance, and its comprehensive accuracy rate is more than 85% within the error range, followed by the nearest neighbor regression algorithm with the accuracy rate reaching 75%, which has a better performance in predicting light rainfall. Both of the above algorithms outperformed the decision tree algorithm with 70% accuracy.

关 键 词：多元线性回归 决策树回归 最近邻回归 降维算法 气象数据质量 

分 类 号：P412[天文地球—大气科学及气象学]