时间序列预处理与信息噪声之间的关系研究——基于离散小波变换和ARIMA模型

Research on the Relationship between Time Series Preprocessing and Information Noise——Based on Discrete Wavelet Transform and ARIMA Model

时间序列的噪声等预处理是数据挖掘及建模过程中重要的一步,对系统分析与预测具有重要意义.基于改良的离散小波变换方法,以2009年5月14日至2019年5月14日为时间范围,以上证指数高频收益率日数据、低频收盘价日数据为实验样本进行去噪预处理,对比四类参数取不同值时的性能表现,并通过ARIMA模型验证预测效果.时间序列预处理与噪声之间不存在矛盾关系,小波方法适当消噪后也可以保留有用信息,提高了分析与预测的正确率.通过研究时间序列预处理与信息噪声之间的关系,期望可以为金融时序的深度挖掘、预测提供一定的指导意见.

[Objective and Significance]The preprocessing of time series noise is an important step in the process of data mining and modeling,and it is of great significance to the analysis and prediction of the system.[Method]Based on the improved discrete wavelet transform method,with the time range from May 14,2009 to May 14,2019,daily data of high-frequency return rate and daily low-frequency closing price of Shanghai Stock Index were used as experimental samples for denoising.Preprocessing,comparing the performance of the four types of parameters with different values,and verifying the prediction effect through the ARIMA model.[Conclusion]There is no contradiction between time series preprocessing and noise,and wavelet method can retain useful information after proper noise reduction,which improves the accuracy of analysis and prediction.This paper studies the relationship between time series preprocessing and information noise,and hopes to provide some guidance for the deep mining and prediction of financial time series.