基于深度经验模态分解的金融市场时序预测

Prediction of time series in financial market based on deep empirical mode decomposition

传统模型和单一模型无法实现时间序列预测的高精度需求,现有时间序列预测模型对一些数据不能做到较为精准的预测。融合经验模态分解(EMD)、主成分分析(PCA)以及长短期记忆网络(LSTM),提出一种深度经验模态分解模型EPL,并提出IEPL(interval EPL)模型进行实验优化。选取4类金融衍生品时间序列的数据集FTSE、S&P500、USD、BDI,以单一模型、传统模型、已有组合模型为对照进行实验。对比实验结果表明,EPL和IEPL在精确度方面表现更好,比现有研究的平均精度提高5%-7%。

The requirements of high precision for time series prediction cannot be realized using traditional models and single models.The existing time series prediction model cannot make predictions for some data more accurately.A deep empirical mode decomposition model(EPL)was proposed by integrating empirical mode decomposition(EMD),principal component analysis(PCA)and long short-term memory networks(LSTM),and IEPL(Interval EPL)model was proposed for experimental optimization.FTSE,S&P500,USD and BDI were selected.Based on the time series datasets of these four types of financial derivatives,comparative experiments were conducted with the single model,the traditional model,and the existing combined model.The results of experiments show that EPL and IEPL perform better in accuracy and improve the average accuracy of existing research by 5%-7%.