基于二次分解重构的股指预测研究

Stock Index Prediction Based on Quadratic Decomposition and Reconstruction

摘要: 针对股票指数复杂难预测的问题，提出一个基于二次分解重构的混合预测模型(ICEEMDAN-EMD- LSTM)。使用改进自适应噪声互补集成经验模态分解(ICEEMDAN)方法将股指序列分解为一系列子序列，并根据模糊熵(FE)的值将这些子序列重构为高频、低频和趋势分量。再使用经验模式分解(EMD)方法对高频分量进行分解，并再次应用FE使高频分量的子序列重构为新的高频、低频和趋势分量。将低频分量、趋势分量和新的高频、低频和趋势分量共五个分量进行线性集成，得到股票指数的最终预测值。为了验证所提出模型的有效性，对沪深300指数序列进行一预测。实验结果表明，与基准模型相比，本文提出的模型方法提高了预测精度，并具有良好的稳健性。

Abstract: A mixed prediction model based on quadratic decomposition and reconstruction (ICEEMDAN-EMD- LSTM) is proposed to address the problem of complex and difficult prediction of stock indices. The improved adaptive noise complementary ensemble empirical mode decomposition (ICEEMDAN) method is used to decompose the stock index sequence into a series of subsequences, and these subsequences are reconstructed into high-frequency, low-frequency, and trend components based on the value of fuzzy entropy (FE). The Empirical Mode Decomposition (EMD) method is used to decompose the high-frequency components, and FE is applied again to reconstruct the subsequences of the high-frequency components into new high-frequency, low-frequency, and trend components. The low-frequency component, trend component, and new high-frequency, low-fre- quency, and trend components are linearly integrated to obtain the final predicted value of the stock index. To verify the effectiveness of the proposed model, a prediction was made on the Shanghai and Shenzhen 300 index series. The experimental results show that compared with the benchmark model, the proposed model method improves prediction accuracy and has good robustness.