兰州站径流支持向量机预测
|
摘要
统计学习理论是研究有限样本情况下机器学习规律的理论。支持向量机是基于统计学习理论的一种新型的机器学习方法,可以解决样本空间中的高度非线性分类和回归等问题。建立了两种模型,模型Ⅰ将黄河干流兰州站的径流时间序列作为输入,模型Ⅱ将径流时间序列和太阳黑子作为输入,两种模型都应用支持向量机对次年的年径流进行预测。结果表明,SVM模型泛化能力强,具有较满意的预测效果。它较好地解决了小样本、过学习、高维数、局部最小等问题,同时模型Ⅱ的预测效果优于模型Ⅰ的,说明径流除了与径流时间序列本身有关外,与太阳黑子活动等有较密切的关系。虽然两者间的物理关系尚需进一步研究,但是支持向量机反映出两者间的非线性关系。
Statistical learning theory is suitable for calculating the statistics of small-sample data.As a new machine learning method based on this theory,the support vector machine(SVM) can solve the problems of nonlinear classification and regression in sample space.This paper develops two SVM models,one takes the runoff of Lanzhou station on Yellow River as its input,and the other takes the same runoff and the sunspot effect as its inputs.Both models are applied to the prediction of the annual runoff of next year.The results show that SVM possesses stronger generalization ability and higher prediction accuracy,and it is a useful tool to overcome the problems of small-sample,over-fitting learning,high-dimension and local minimum trapping.An improvement of prediction results by the second model over the first one indicates that the runoff is closely related to the sunspot activity.Though the physical meaning of this relation is not clear yet,the nonlinear relationship can be modeled by the SVM model.
Statistical learning theory is suitable for calculating the statistics of small-sample data.As a new machine learning method based on this theory,the support vector machine(SVM) can solve the problems of nonlinear classification and regression in sample space.This paper develops two SVM models,one takes the runoff of Lanzhou station on Yellow River as its input,and the other takes the same runoff and the sunspot effect as its inputs.Both models are applied to the prediction of the annual runoff of next year.The results show that SVM possesses stronger generalization ability and higher prediction accuracy,and it is a useful tool to overcome the problems of small-sample,over-fitting learning,high-dimension and local minimum trapping.An improvement of prediction results by the second model over the first one indicates that the runoff is closely related to the sunspot activity.Though the physical meaning of this relation is not clear yet,the nonlinear relationship can be modeled by the SVM model.
引文
[1]陈洪滨,范学花.2007年极端天气和气候事件及其他相关事件的概要回顾[J].气候与环境研究,2008,13(1):102～112.CHEN Hongbin,FAN Xuehua.Some extreme events of weather,climate and related phenomena in 2007[J].Climatic andEnvironmental Research,2008,13(1):102～112.(in Chinese)
[2]赵海娟,梁红飞,占腊生,等.预报第23周平滑月平均太阳黑子数的一种方法[J].天文学报,2003,44(3):279～285.ZHAO Haijuan,LIANG Hongfei,ZHAN Lasheng,et al.A method for the prediction of the monthly smoothed sunspot numbers of the23th cycle[J].Acta Astronomica Sinica,2003,44(3):279～285.(in Chinese)
[3]赵佩章,陈健,赵文桐.太阳黑子对厄尔尼诺、拉尼娜的影响[J].地球物理学进展,2001,16(3):85～90.ZHAO Peizhang,CHEN Jian,ZHOU Wentong.The sunspots affect Einino and lanina[J].Progress in Geophysics,2001,16(3):85～90.(in Chinese)
[4]陈文玉,金德山,李培,等.太阳黑子对南海季风爆发及热带风暴活动的影响[J].热带海洋,2000,19(4):23～27.CHEN Wenyu,JIN Deshan,LI Pei,et al.Effects of sunspot on eruption of south China sea monsoon and activity of tropical storm[J].Tropic Oceanology,2000,19(4):23～27.(in Chinese)
[5]Vapnik V.The nature of statistical learning theory[M].New York:Springer,1995.
[6]Nello C,John S T.An introduction to support vector machines and other kernel-based learning methods[M].London:CambridgeUniversity Press,2000.
[7]Shevade S K,Keerthi S S.Bhattacharyy C,et al.Improvements to SMO algorithm for SVM regression[J].IEEE Trans on NauralNetworks,2000,11(5):1188～1193.
[8]王炜,刘悦,李国正,等.中国大陆强震时间序列预测的支持向量机方法[J].地震,2005,25(4):26～32.WANG Wei,LIU Yue,LI Guozheng,et al.The support vector machine method for time sequence forecasting of strong earthquakes inChina′s mainland.Earthquake,2005,25(4):26～32.(in Chinese)
[9]Bond G,Showe W,Cheseby M,et al.A pervasive millennial scale cycle in North Atlantic Holocene and glacial climates[J].Science,1997,278(5341):1257～1266.
[10]吴燕辉,周勇.1958-2004年江汉平原气温变化特征分析[J].气象与环境学报,2007,23(3):5～9.WU Yanhui,ZHOU Yong.Characteristics of air temperature changes in Jianghan plain from 1958 to 2004[J].Journal of Methorologyand Environment,2007,23(3):5～9.(in Chinese)
[2]赵海娟,梁红飞,占腊生,等.预报第23周平滑月平均太阳黑子数的一种方法[J].天文学报,2003,44(3):279～285.ZHAO Haijuan,LIANG Hongfei,ZHAN Lasheng,et al.A method for the prediction of the monthly smoothed sunspot numbers of the23th cycle[J].Acta Astronomica Sinica,2003,44(3):279～285.(in Chinese)
[3]赵佩章,陈健,赵文桐.太阳黑子对厄尔尼诺、拉尼娜的影响[J].地球物理学进展,2001,16(3):85～90.ZHAO Peizhang,CHEN Jian,ZHOU Wentong.The sunspots affect Einino and lanina[J].Progress in Geophysics,2001,16(3):85～90.(in Chinese)
[4]陈文玉,金德山,李培,等.太阳黑子对南海季风爆发及热带风暴活动的影响[J].热带海洋,2000,19(4):23～27.CHEN Wenyu,JIN Deshan,LI Pei,et al.Effects of sunspot on eruption of south China sea monsoon and activity of tropical storm[J].Tropic Oceanology,2000,19(4):23～27.(in Chinese)
[5]Vapnik V.The nature of statistical learning theory[M].New York:Springer,1995.
[6]Nello C,John S T.An introduction to support vector machines and other kernel-based learning methods[M].London:CambridgeUniversity Press,2000.
[7]Shevade S K,Keerthi S S.Bhattacharyy C,et al.Improvements to SMO algorithm for SVM regression[J].IEEE Trans on NauralNetworks,2000,11(5):1188～1193.
[8]王炜,刘悦,李国正,等.中国大陆强震时间序列预测的支持向量机方法[J].地震,2005,25(4):26～32.WANG Wei,LIU Yue,LI Guozheng,et al.The support vector machine method for time sequence forecasting of strong earthquakes inChina′s mainland.Earthquake,2005,25(4):26～32.(in Chinese)
[9]Bond G,Showe W,Cheseby M,et al.A pervasive millennial scale cycle in North Atlantic Holocene and glacial climates[J].Science,1997,278(5341):1257～1266.
[10]吴燕辉,周勇.1958-2004年江汉平原气温变化特征分析[J].气象与环境学报,2007,23(3):5～9.WU Yanhui,ZHOU Yong.Characteristics of air temperature changes in Jianghan plain from 1958 to 2004[J].Journal of Methorologyand Environment,2007,23(3):5～9.(in Chinese)
版权所有：© 2023 中国地质图书馆 中国地质调查局地学文献中心