摘要
混沌时间序列预测研究的2个焦点:一个是增加预测模型的复杂度,以面向控制、水文、气象,脑电生理学等研究背景下的具体预测需求;另一个是引入和改进模式识别领域里的特征提取算法,从而降低混沌数据的预测难度,以提高预测精度。采用经验模态分解和独立成分分析算法,改进线性和非线性特征的提取。并在解析意义下,给出了一种新颖的隐式特征表达。在不改进预测模型的前提下,提出了一种混沌序列隐式特征提取算法。采用经典的Mackey-Glass仿真、比利时皇家天文台太阳黑子数,以及密西西比河实测流量数据实验表明,该方法提高了模型预测精度。
Chaotic time series prediction is an active research area and has received considerable attention. Previous research has focused on two aspects, one is improving the forecasting model complexity to meet the requirements of the applications in variety of areas, such as control, hydrology, meteorology and cerebral electrophysiology, the other is introducing and improving the feature extraction algorithm in pattern recognition field. The aim of this work is to de- crease the prediction difficulty of chaotic data and improve the prediction accuracy. This paper adopts empirical mode decomposition and independent component analysis algorithms to improve the extraction of linear and non-linear fea- tures;and a novel hidden feature expression is given in the sense of analysis. A feature extraction method in hidden mode for chaotic time series prediction is proposed without improving the prediction model. The results of classical Mackey-Glass time series simulation and the experiments on sunspot data from Royal Observatory of Belgium and Mississippi river flow time series show that the forecasting results of the proposed hidden feature extraction method are superior to those of Elman neural network, least squares support vector machine and the Elman neural network com- bined with empirical mode decomposition.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2014年第1期1-7,共7页
Chinese Journal of Scientific Instrument
基金
教育部新世纪优秀人才计划(NCET10-0062)资助项目
关键词
混沌时间序列
隐式特征提取
经验模态分解
独立成分分析
chaotic time series
hidden feature extraction
empirical mode decomposition
independent component analysis