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.
出处
《数学的实践与认识》
北大核心
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)
关键词
日极端气温
降雨
函数型数据
高斯过程
预测
daily extreme air temperature
rainfall
functional data
gaussian process
prediction.
作者简介
通讯作者
