基于CEEMDAN-VMD-LSTM的超高频金融时间序列预测  

UHF financial time series predicting based on CEEMDAN-VMD-LSTM

作　　者：闫勇志 沐年国[1] Yan Yongzhi;Mu Nianguo(Business School,University of Shanghai for Science&Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学管理学院,上海200093

出　　处：《计算机时代》2023年第5期102-108,共7页Computer Era

摘　　要：对超高频金融数据的预测,模态分解降低了数据的噪声,提高了数据预测精度。据此提出了自适应噪声的完整集合经验模态分解(CEEMDAN)与变分模态分解(VMD)相结合的二次分解模型。先将期货日度行情数据通过CEEMDAN一次分解,并通过样本熵将分解后的序列整合成高频、低频和趋势序列;再将高频和低频序列分别进行VMD分解,然后将各个IMF分量通过LSTM网络预测,最终整合各个预测结果。模型各项指标均好于一次分解。For the prediction of UHF financial data,modal decomposition reduces the noise of data and improves the accuracy of data prediction.Accordingly,a quadratic decomposition model combining the complete ensemble empirical mode decomposition for adaptive noise(CEEMDAN)and variational mode decomposition(VMD)is proposed.Firstly,the daily market data of futures are decomposed by CEEMDAN,and the decomposed series are integrated into high-frequency,low-frequency and trend series by sample entropy.Then,the high-frequency and low-frequency series are decomposed by VMD respectively,and each IMF component is predicted by LSTM network.Finally,the prediction results are integrated.Compared with CEEMDAN decomposition and VMD decomposition,each index of the proposed model is better.

关 键 词：超高频金融数据 CEEMDAN VMD 二次分解 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]