基于CEEMDAN和LSTM的股指价格组合预测方法:来自5个国家的数据分析  

A Combinatorial Prediction Method for Stock Index Price Based on CEEMDAN and LSTM: Data Analysis from Five Countries

作　　者：林昇铭 王嘉宏 袁锦琛 曾莹萍 林持旺 

机构地区：[1]福建理工大学计算机科学与数学学院,福建 福州 [2]福建理工大学大数据挖掘与应用技术重点实验室,福建 福州

出　　处：《计算机科学与应用》2024年第2期449-459,共11页Computer Science and Application

摘　　要：股票价格预测一直是研究者们挑战的领域。然而,现有的基于深度学习的预测方法在预测不同国家的股票指数时无法达到统一优秀的效果。因此,本文提出了一种名为CEEMDAN-PSO-LSTM的新模型来预测多个国家的股票指数收盘价。首先,我们使用完全自适应噪声集合经验模态分解(CEEMDAN)方法将原始股票指数收盘价序列分解为多个本征模态函数(IMF)。然后,得到的各个IMF通过利用粒子群优化算法(PSO)优化长短期记忆网络(LSTM)的超参数后的模型进行预测,最终,将各IMF的预测结果进行加和得到对原始收盘价序列的预测结果。为验证所提方法的可行性,我们将其与LSTM、PSO-LSTM、EMD-LSTM和CEEMDAN-LSTM这四个模型进行对比,并选取五支来自不同国家且具有代表性的股票指数作为数据集。通过各模型在各数据集上的实验表明本文所提方法表现优于其他四个模型,这说明本文所提方法具有优良的可行性和普适性。Stock price prediction has always been a challenging field for researchers. However, existing deep learning based prediction methods cannot achieve uniform and excellent results in predicting stock indices of different countries. Therefore, this article proposes a new model called CEEMDAN- PSO-LSTM to predict the closing prices of stock indices in multiple countries. Firstly, we use the complete ensemble empirical mode decomposition of adaptive noise (CEEMDAN) method to de-compose the original stock index closing price sequence into multiple intrinsic mode functions (IMF). Then, the obtained IMF models are predicted by using the particle swarm optimization (PSO) algorithm to optimize the hyperparameters of the Long Short-Term Memory Network (LSTM). Finally, the prediction results of each IMF are summed to obtain the prediction results of the original closing price series. To verify the feasibility of the proposed method, we compared it with four mod-els: LSTM, PSO-LSTM, EMD-LSTM, and CEEMDAN-LSTM, and selected five representative stock indices from different countries as the dataset. The experiments of each model on various datasets show that the proposed method performs better than the other four models, indicating that the proposed method has excellent feasibility and universality.

关 键 词：长短期记忆网络 完全自适应噪声集合经验模态分解 粒子群优化算法 股票价格预测 深度学习 

分 类 号：F83[经济管理—金融学]