融合支持向量机算法的数据驱动型数据同化方法研究
Data-driven Data Assimilation Method based on Support Vector Machine Algorithm
余青何 1摆玉龙 1范满红1
作者信息
- 1. 西北师范大学 物理与电子工程学院,甘肃 兰州 730070
- 折叠
摘要
数据驱动建模是从数据中探究状态变量的时空演化关系.数据驱动型数据同化方法是探索使用数据驱动模型替代传统(基于物理的)模型,实现优化融合观测信息与模型模拟的同化方法.研究将数据驱动的支持向量机回归预测模型应用于集合卡尔曼滤波过程中,使用模拟预测方法对动力学系统进行非参数采样得到系统轨迹的代表性样本集,从样本集中重构动力学系统.提出一种支持向量机回归机器学习模拟预测策略的数据驱动数据同化方法,并将其应用于经典模式驱动同化系统.采用Lorenz-63和Lorenz-96非线性模型进行数值实验.通过改变样本集大小、噪声方差和观测步长等敏感性参数比较数据同化性能.结果表明:对于较大的样本集,该组合方法优于一般的顺序数据同化方法,从而证明新方法的有效性.
Abstract
Data-driven modeling is to discover the spatio-temporal evolution of state variables from data.Data-driven data assimilation is a scientific method to optimize the fusion of observation information and model by us-ing data-driven model instead of traditional(physics-based)model.In this work,a data-driven support vector machine regression prediction model is applied to the ensemble Kalman filtering process,and the dynamic sys-tem is reconstructed from the sample set by non-parametric sampling of the dynamic system trajectory using the simulation prediction method.A data driven data assimilation method based on support vector machine regres-sion machine learning simulation prediction strategy is proposed and applied to classical pattern driven data as-similation system.The Lorenz-63 and Lorenz-96 model are used for numerical experiments.The data assimila-tion performance is compared by changing the sensitivity parameters such as sample sizes,noise variance and ob-servation step sizes.The results show that the proposed method is superior to the general sequential data assimi-lation method for large sample sets,which proves the effectiveness of the new method.
关键词
数据驱动数据同化/支持向量机/集合卡尔曼滤波/Lorenz模型Key words
Data-driven data assimilation/Support vector machine/Ensemble Kalman filter/Lorenz model引用本文复制引用
出版年
2024
NETL
NSTL
万方数据