基于支持向量回归的海面高度统计预报方法研究  

Statistical forecast for sea surface height based on support vector regression

作　　者：杨毅 王亚波[1] 邓林 杨宗元[1] 袁克非[1] 杨建[1] YANG Yi;WANG Yabo;DENG Lin;YANG Zongyuan;YUAN Kefei;YANG Jian(Wuhan Second Ship Design and Research Institute,Wuhan 430010,China)

机构地区：[1]武汉第二船舶设计研究所,湖北武汉430010

出　　处：《海洋通报》2022年第5期548-555,共8页Marine Science Bulletin

摘　　要：本文提出了一种基于支持向量回归的统计预报方法,通过经验正交分解对原始数据矩阵进行时空分解,提取出空间模态和时间系数。由于海面高度变化具有非线性、大惯性的特点,对时间系数进行小波分析,能有效滤除其中的高频信号,得到表征海面高度变化的低频信号。利用支持向量回归方法对小波分解后的低频信号构建预报模型,然后进行小波重构,还原时间序列长度,实现未来7天的海面高度预报。通过黑潮附近海域的海面高度预报结果进行验证,该预报方法的预报效果优于整合滑动平均自回归预报方法。本文通过机器学习的算法实现了海面高度的预报,为海洋预报方法提供了新的思路。This paper proposes a statistical forecasting method based on support vector regression,which decomposes the sea surface height data with empirical orthogonal functions for extracting spatial patterns and temporal coefficients.The wavelet analysis of the time coefficients is performed to effectively filter out the high-frequency signals and obtain the lowfrequency signals representing the sea surface height change,in response to the nonlinear changes of sea surface height.The low-frequency signals from wavelet analysis are modeled by support vector regression,and the time series are reverted with wavelet reconstruction to achieve sea surface height forecasts for the next 7 days.The results of sea surface height forecasting in the sea near Kuroshio demonstrate that the proposed method is superior to the Autoregressive integrated moving average method.This prediction of sea surface height with machine learning algorithm provides a novel approach to ocean forecasting.

关 键 词：统计预报 海面高度异常 小波分析 经验正交分解 支持向量回归 

分 类 号：P731.34[天文地球—海洋科学]