摘要
利用浙江省杭州和椒江两站2003—2005年的晴空逐时气象数据,采用最小二乘支持向量机方法（LS-SVM）,建立了晴空逐时太阳辐射模型.模型的输入因子为逐时天文辐射、气温、气压、水汽压、能见度和风速等要素,输出因子为逐时太阳辐射.模型数据分为2部分,其中2003年的数据用于训练建模,2004—2005年的数据用于模型的评估.结果表明,LS-SVM方法能够很好地模拟气象要素对太阳辐射的非线性影响,建立的太阳辐射模型精度较高,模型的解释性方差R2为0.950 5,均方根误差ERMS=0.159 0 MJ·m^-2,平均误差EMB和平均绝对误差EMAB分别为0.005 3和0.124 1 MJ·m^-2.根据浙江省68个气象站2005-12-15T14：00的气象要素,估算出的太阳辐射为1.39~2.24 MJ·m^-2.基于LS-SVM方法的晴空逐时太阳辐射模型具有很好的学习推广能力,利用常规的气象观测资料,即可模拟出具有相似气候背景下的晴空逐时太阳辐射,为太阳辐射遥感反演提供地面数据.
Remote sensing is the most feasible way to study global or regional solar radiation. However, due to the scarcely available in-situ data, the validation of retrieval solar radiation is hard to perform. The least squares support vector machines （LS-SVM） is utilized to estimate the hourly solar radiation in clear days. The input parameters of the LS-SVM model include hourly extraterrestrial radiation, temperature, pressure, water vapor pressure, visibility and wind speed. The hourly meteorological data of Hangzhou and Jiaojiang from year 2003 to 2005 are split into two parts. The data in 2003 are used to train the model, and data in 2004-2005 are used to test the model. The result indicates that SVM model performs satisfactorily： Rz =0. 950 5, ERMS= 0. 159 0 MJ·m^-2, EMB=0. 005 3 MJ·m^-2, and EMAB =0. 124 1 MJ·m^-2. The LS-SVM solar radiation model has good ability of modeling nonlinear process and advantage of generality. We input routine meteorological data of 68 weather stations in Zhejiang into the LS-SVM model and estimate the solar radiation at 2005-12-15T14.00 （ranging from 1.39 to 2.24 MJ·m^-2）. The results demonstrate that we can achieve accurate solar radiation by using LS-SVM when in-situ solar radiation data is missing or not available.
出处
《北京师范大学学报（自然科学版）》
CAS
CSCD
北大核心
2007年第3期274-278,共5页
Journal of Beijing Normal University(Natural Science)
基金
国家自然科学青年科学基金资助项目(40601067)
中国科学院知识创新工程重要方向资助项目(KZCXZ-YW-313)
关键词
太阳辐射
最小二乘支持向量机
模拟技术
solar radiation
least squares support vector machines
simulation technology
