基于二次分解与LSTM的金融时间序列预测算法研究

现有结合特征提取与预测模型的方法不能准确把握金融时间序列的混沌性与交互性，导致预测精度不高。针对此问题，提出一种基于二次分解与长短期记忆(long short term memory, LSTM)网络的金融时间序列预测算法。使用变分模态分解方法与集成经验模态分解方法依次解析金融时间序列数据，得到能表达数据混沌性特征的模态;将模态信息输入到融合有因子分解机(factorization machine, FM)的长短期记忆网络模型中，融合获取到的长记忆性特征与交互性特征，进而预测最终的结果;选取沪深300指数的历史数据作为实验数据集，通过多组对比实验验证算法的有效性。实验结果表明,提出的算法可以有效提升模型的预测能力，同时表达金融时间序列的混沌性、长记忆性、交互性。

Research on financial time series forecasting algorithm based on secondary decomposition and LSTM

The existing methods combining feature extraction and prediction model cannot accurately grasp the chaos and interaction of financial time series, resulting in low prediction accuracy. A financial time series prediction algorithm based on secondary decomposition and long short term memory (LSTM) is proposed to solve this problem. The variational mode decomposition method and the ensemble empirical mode decomposition method are used to analyze the financial time series data, so as to obtain the mode that can express the chaotic characteristics of the data. Then the modal information is input into the long-term and short-term memory network model fused with factorization machine (FM), the obtained long-term memory features and interactive features are fused, and then the final result is predicted. The historical data of CSI 300 index is selected as the experimental data set, and the effectiveness of the algorithm is verified by multiple groups of comparative experiments. Experimental results show that the proposed algorithm can effectively improve the prediction ability of the model and express the chaos, long memory and interaction of financial time series.