Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (5): 239-244.DOI: 10.3778/j.issn.1002-8331.2009-0186

Previous Articles     Next Articles

Prediction of Financial Time Series by EMD-LSTM Model

YAO Honggang, MU Nianguo   

  1. School of Management, University of Shanghai for Science and Technology, Shanghai 200093, China
  • Online:2021-03-01 Published:2021-03-02

EMD-LSTM模型对金融时间序列的预测

姚洪刚,沐年国   

  1. 上海理工大学 管理学院,上海 200093

Abstract:

Aiming at the high noise and nonlinear characteristics of financial time series, this paper proposes a financial time series prediction model based on Empirical Mode Decomposition(EMD) and Long Short-Term Memory(LSTM) network. In order to avoid using the test set information in the model training process after only one empirical mode decomposition of the overall sequence, the time series data is subjected to multi-step empirical mode decomposition through a certain size time window. Next, the denoising and reconstruction of the decomposed sequence is performed, and then the reconstructed sequence is used as the input of the LSTM network to obtain the final prediction result.Using SSE composite index data,comparing it with the standard LSTM model and common prediction methods combined with EMD, the results show that the proposed EMD-LSTM model has better prediction effects.

Key words: financial time series, empirical mode decomposition, long short-term memory network, time window

摘要:

针对金融时间序列高噪声以及非线性的特点,提出一种基于经验模态分解(EMD)和长短期记忆(LSTM)网络的金融时间序列预测模型。为避免对整体序列只进行一次经验模态分解后的模型训练过程中使用测试集的信息,将时间序列数据通过一定大小的时间窗口进行多步经验模态分解,并对分解后的序列去噪重构,再将重构后的序列作为LSTM网络的输入,得到最终的预测结果。利用上证综指数据,将其与标准LSTM模型以及常见的结合EMD的预测方法进行对比,结果表明提出的EMD-LSTM模型具有更好的预测效果。

关键词: 金融时间序列, 经验模态分解, 长短期记忆网络, 时间窗口