基于高斯过程函数型回归模型的日最高气温研究  

On Research of Daily Highest Air Temperature based on Gaussian Regression Functional Regression Model

作　　者：张博[1] 曹春正[1] 朱晓欣[1] 朱连华[1] 

机构地区：[1]南京信息工程大学数学与统计学院,江苏南京210044

出　　处：《数学的实践与认识》2016年第4期181-189,共9页Mathematics in Practice and Theory

基　　金：国家自然科学基金(11301278);教育部人文社科基金(13YJC910001);江苏省自然科学基金(BK2012459)

摘　　要：极端天气是目前社会热点问题.利用高斯过程函数型回归对北京,上海等10个城市近年来夏季日最高气温进行整体建模.选取城市地理位置信息作为均值函数解释变量,时间和降雨信息作为高斯过程协方差结构解释变量,充分利用模型能够同时捕捉均值和协方差结构的优势,解决多地区日最高气温的整体建模和同步预测问题.研究表明,高斯过程函数型回归模型在随机预测,外延预测,k步预测,以及对于训练数据集以外城市的预测均有较好的效果,且优于一般的函数型数据模型.Extreme weather is a hot social issue.In this paper,Gaussian process regression functional model is considered for modeling the daily highest temperature in recent summers on Beijing,Shanghai and other ten cities in China.The urban location information is used as explanatory variable of the mean function,while time and rainfall information are used as explanatory variables of the covariance structure in the Gaussian process.The model can capture both the mean and covariance structure and forecast the daily highest temperature on different cities synchronously.Studies have shown that the Gaussian process functional regression model had good prediction results on random,extension and k step prediction,as well as for a city outside the training data,and also is superior to some general functional data models.

关 键 词：日极端气温 降雨 函数型数据 高斯过程 预测 

分 类 号：P423.36[天文地球—大气科学及气象学] O212.1[理学—概率论与数理统计]