基于CEEMDAN与LSTM-Attention的股市预测模型

THE STOCK MARKET PREDICTION MODEL BASED ON CEEMDAN AND LSTM-ATTENTION

具有时序特征的金融股票数据有非线性、非平稳和复杂动态的特点,对预测模型提出了挑战。提出一种基于自适应噪声完备集合经验模态分解的LSTM-Attention模型。通过重组后的高频、中频和低频分量,构建更为细化的LSTM-Attention模型,再通过加总合成获得目标预测值。实验结果分析表明,该模型在平均绝对误差(MAE)、均方根误差(RMSE)、均方误差(MSE)和决定系数四个指标上均优于现有模型,有效提升了模型预测的准确率,同时减少了计算开销。

Financial stock data with time-series characteristics are non-linear,non-stationary and complex dynamic,which poses a challenge to the prediction model.This paper proposes the LSTM-Attention model based on adaptive noise complete set of empirical mode decomposition.By restructuring after the high,medium and low frequency components,a more refined LSTM-Attention model was built.Target forecast was obtained by aggregation integration.The analysis of the experimental results shows that the mean absolute error(MAE),the root mean square error(RMSE),mean square error(MSE)and decision coefficient of this method are better than those of existing models,and it effectively improves the model prediction accuracy,and reduce the computational overhead.