地磁变化场的MEEMD-样本熵-LSSVM预测模型
Forecasting Model of Geomagnetic Variation Field Based onModified Ensemble Empirical Mode Decomposition-SampleEntropy-Least Square Support Vector Machine
-
摘要: 目的 针对地磁变化场时间序列的混沌特性,提出了一种改进的集成经验模态分解(modified ensemble em-pirical mode decomposition,MEEMD)-样本熵-最小二乘支持向量机(least square support vector machine,LSSVM)的地磁变化场预测模型。首先,利用MEEMD-样本熵将非平稳的地磁变化场时间序列分解为一系列复杂度差异明显的地磁变化场子序列;然后,针对每一个子序列分别建立LSSVM模型,选择各自适合的最优模型参数;最后,以地磁台站实测的地磁变化场数据为例进行实验,并与基于单一LSSVM以及RBF径向基神经网络的两种预测模型进行比较。实验结果表明,MEEMD-样本熵-LSSVM模型的预测值能紧跟地磁变化场的变化趋势,相比另外两种模型,体现出更好的预测效果,在地磁Kp指数小于3时,预测3h平均绝对误差为1.63nT。Abstract: Objective Modeling and forecasting of the geomagnetic variation field is the important research topic ofgeomagnetic navigation and space environment monitoring.According to the chaotic feature of geo-magnetic variation time series,a combined forecasting model based on modified ensemble empiricalmode decomposition(MEEMD)-sample entropy(SampEn)-least square support vector machine(LSS-VM)is proposed.Firstly,the geomagnetic variation time series is decomposed into a series of geo-magnetic variation subsequences with obvious differences in complex degree using MEEMD-SampEn.Then,the forecasting model of each subsequence is created with LSSVM using the optimal model pa-rameters.Finally,the simulation is performed by using the real data collected from the geomagneticobservatory.The results show that the forecasting value of the MEEMD-SampEn-LSSVM model canclosely keep up with the trend of geomagnetic variation field,and obviously better than the other twomodels.The mean absolute error of the model forecasting three hours is 1.63nT when Kplessthan 3.
下载:
