基于最小二乘支持向量机的副热带高压预测模型
Subtropical High Forecast Model of Least Square Support Vector Machine
-
摘要: 采用EOF时空分解、小波频牢分解和最小二乘支持向量机(LS-SVM)交叉互补方法,建立夏季500 hPa位势高度场的预测模型,用以描绘和表述副热带高压形势场的形态和变化。首先用经验正交函数分解(EOF)方法将NCEP/NCAR再分析资料500 hPa位势高度场序列分解为彼此正交的特征向量及其对应时间系数,随后提取前15个主要特征向龟的时间系数(方差贡献96.2%),采用小波分解方法将其分解为相对简单的带通信号,再利用LS-SVM方法建立各分量信号的预测模型,最后通过小波时频分量重构和EOF时空重构,得到500hPa位势高度场的预测结果以及副热带高压形势场的预测。通过对预测模型的试验情况和分析对比,结果表明:基于上述思想提出的算法模型能较为准确地描述500 hPa位势高度场的形态分布并预测1~7 d的副热带高压活动,对10~15 d的副热带高压活动预测结果也有参考意义。Abstract: Based on the methods of empirical orthogonal decomposition (EOF), wavelet frequency decomposition and least square support vector machine, a summer 500 hPa potential height forecasting model is established to describe the form and change of the subtropical high situation. First, 500 hPa potential height fields sequences on NCEP/NCAR are separated into the time coefficients and corresponding eigenvectors which are orthogonal to each other with the method of empirical orthogonal decomposition. Then fifteen time coefficient series corresponding with major eigenvector (square contribution of 96.2%) are extracted and each time coefficient is decomposed to relatively simple signals with the method of wavelet analysis. Then, each signal prediction model is set up with the method of least square support vector machine. Finally, the forecasting simple signals are used to reconstruct the corresponding forecasting time series with the method of wavelet decomposition, then, the forecasting time series and corresponding major eigenvector are used to reconstruct 500 hPa potential fields with the methods of empirical orthogonal decomposition. The reconstructed potential fields are the fields which are forecasting results. Through experiments and analysis of contrast on the prediction model, the results show that the proposed algorithm model based on the above ideas can basically describe the distribution of 500 hPa potential situation and basically forecast the location and intensity of subtropical high within seven days. And the results also show that the 10-15 day forecasting results by the model can be used for reference for the medium and long-term activity of the subtropical high. The results also show that the model exhibits its properties of simplicity, stability, flexibility and good prospect of application.
-
图 1 实况与预测对比试验(单位:gpm)
(a) 2004年8月15日500 hPa高度场实况,(b) 1 d预测位势场,(c) 3 d预测位势场,(d) 5 d预测位势场,(e) 7 d预测位势场,(f) 10 d预测位势场,(g) 15 d预测位势场
Fig. 1 Observation and prediction comparative test (unit:gpm)
(a) 500 hPa potential field on August 15, 2004, (b) 1-day prediction potential field, (c) 3-day prediction potential field, (d) 5-day prediction potential field, (e) 7-day prediction potential field, (f) 10-day prediction potential field, (g) 15-day prediction potential field
表 1 2003—2005年5月1日一8月31日预测位势场和实况场的平均相关系数及均方差
Table 1 Average correlation coefficients and variances between prediction and observation fields from May 1 to Augast 31 in 2003-2005
-
[1] 张韧.基于前传式网络逼近的太平洋副热带高压活动的诊断预测.大气科学, 2001, 25(5):649-660 http://www.cnki.com.cn/Article/CJFDTOTAL-DQXK200105006.htm [2] Zhang Ren, Yu Zhihao. Neural network BP model approximation and prediction of complicated weather systems. ACTA Meteorologica Sinica, 2001, 15(1):105-115 http://www.cnki.com.cn/Article/CJFDTotal-QXXW200101010.htm [3] Vapink V, Goloeich S, Smola A. Support Vector Method for Function Approximation, Regression Estimation, and Signal Processing. Cambridge:MIT Press, 1997:281-287 [4] Smola A J. Regression Estimation with Support Vector Learning Machines. Technische University at Myunchen, 1996 [5] 马云潜, 张学工.支持向量机函数拟合在分形插值中的应用.清华大学学报, 2000, 40(3):76-78 http://www.cnki.com.cn/Article/CJFDTOTAL-QHXB200003019.htm [6] 董辉, 傅鹤林, 冷伍明.支持向量机的时间序列回归与预测.系统仿真学报, 2006, 18(7):1785-1788 http://www.cnki.com.cn/Article/CJFDTOTAL-XTFZ200607014.htm [7] 林剑艺, 程春田.支持向量机在中长期径流预报中的应用.水利学报, 2006, 37(6):681-686 http://www.cnki.com.cn/Article/CJFDTOTAL-SLXB200606006.htm [8] 陈永义, 俞小鼎, 高学浩, 等.处理非线性分类和回归问题的一种新方法 (Ⅰ)——支持向量机方法简介.应用气象学报, 2004, 15(3):345-354 http://qikan.camscma.cn/jams/ch/reader/view_abstract.aspx?file_no=20040344&flag=1 [9] 冯汉中, 陈永义.处理非线性分类和回归问题的一种新方法 (Ⅱ)——支持向量机方法在天气预报中的应用.应用气象学报, 2004, 15(3):355-365 http://qikan.camscma.cn/jams/ch/reader/view_abstract.aspx?file_no=20040345&flag=1 [10] 刘科峰, 张韧, 万齐林等.结构风险极小的支持向量机方法及其在副热带高压数值预报优化中应用.应用基础与工程科学学报, 2006, 14:384-390 [11] 刘科峰, 张韧, 徐海斌, 等.支持向量机与卡尔曼滤波集合的西太平洋副热带高压数值预报误差修正.气象学报, 2007, 65(3):141-157 http://www.cnki.com.cn/Article/CJFDTOTAL-QXXB200703012.htm [12] Suykens J A K, Vandewalle J. Least squares support vector machine classifiers. Neural Processing Letters, 1999, 9(3):293-300 doi: 10.1023/A:1018628609742 [13] Suykens J A K, Van Gestel T, de Brabanter J, et al. Least Squares Support Vector Machines. Singapore:World Scientific Pub Co, 2002 [14] 张韧, 何金海, 董兆俊.南亚夏季风影响西太平洋副高南北进退活动的小波包能量诊断.热带气象学报, 2004, 20(2):113-121 http://www.cnki.com.cn/Article/CJFDTOTAL-RDQX200402000.htm [15] 朱家元, 杨云, 张恒喜.基于优化最小二乘支持向量机的小样本预测研究.航空学报, 2004, 25(6):565-568 http://www.cnki.com.cn/Article/CJFDTOTAL-HKXB200406008.htm [16] 张学工.关于统计学习理论与向量机.自动化学报, 2000, 26(1):32-42 http://www.cnki.com.cn/Article/CJFDTOTAL-MOTO200001005.htm [17] 刘科峰, 张韧.EOF分解与Kalman滤波相结合的副高位势场数值预报优化.解放军理工大学学报 (自然科学版), 2006, 7(3):291-296. http://www.cnki.com.cn/Article/CJFDTOTAL-JFJL200603017.htm