Abstract: In order to improve the prediction accuracy of financial series, a combined EMD-SSA-LSTM-SVR prediction model based on empirical modal decomposition（EMD） and singular spectrum analysis（SSA） is proposed. The model combines the respective advantages of EMD decomposition and SSA decomposition, decomposes the original financial series into components with different time scales, and then makes full use of the advantages of the LSTM model in dealing with long-term dependent series and the generalization ability of the SVR model for non-linear series to forecast each component, and finally integrates them to obtain the predicted values of the financial series. Experiments show that the EMD-SSA-LSTM-SVR model has higher prediction accuracy than existing EMD and SSA-based prediction models such as EMD-LSTM, EMD-SVR, SSA-SVR and SSA-LSTM.

Key words: empirical modal decomposition（EMD）, singular spectrum analysis（SSA）, long short-term memory networks（LSTM）, support vector regression（SVR）

摘要： 为了提高金融序列的预测精度，提出了一种基于经验模态分解（EMD）和奇异谱分析（SSA）的EMD-SSA-LSTM-SVR组合预测模型。该模型结合了EMD分解和SSA分解各自的优点，将原始金融序列分解为具有不同时间尺度的分量，充分发挥LSTM模型处理长期依赖序列的优势以及SVR模型对非线性序列的泛化能力对各个分量进行预测，集成得到金融序列的预测值。实验表明，与现有的EMD-LSTM、EMD-SVR、SSA-SVR和SSA-LSTM等基于EMD和SSA的预测模型相比，EMD-SSA-LSTM-SVR模型具有更高的预测精度。

关键词: 经验模态分解, 奇异谱分析, 长短时记忆网络, 支持向量回归