基于VMD-PCA-BiLSTM的股票价格预测

Stock Price Prediction Based on VMD-PCA-BiLSTM

摘要: 针对股票时间序列非平稳、高噪声导致预测效果受限的问题，本文提出VMD-PCA-BiLSTM股票价格预测模型。结合平均样本熵构建自适应变分模态分解，对收盘价序列做分解处理，实现频域降噪与多尺度特征解耦；同时采用主成分分析，对量价数据及技术指标进行降维，消除变量多重共线性，重构低维时域特征空间。模型搭建双分支双向长短期记忆网络，分别提取频域、时域特征并深度融合以完成预测。实验表明，该模型有效改善传统方法的平滑效应与时滞缺陷，均方根误差降至28.55，决定系数达97.95%，预测精度与鲁棒性提升显著。

Abstract: To address the limitations in forecasting performance caused by the non-stationary nature and high noise levels of stock time series, this paper proposes the VMD-PCA-BiLSTM stock price forecasting model. By combining average sample entropy with adaptive variational mode decomposition, the model decomposes closing price sequences to achieve frequency-domain denoising and multi-scale feature decoupling. Simultaneously, principal component analysis is employed to reduce the dimensionality of volume-price data and technical indicators, eliminating multicollinearity among variables and reconstructing a low-dimensional time-domain feature space. The model employs a dual-branch bidirectional long short-term memory (BiLSTM) network to extract frequency-domain and time-domain features, respectively, and deeply integrate them to perform forecasting. Experimental results demonstrate that this model effectively addresses the smoothing effects and time lag issues inherent in traditional methods, reducing the root mean square error (RMSE) to 28.55 and achieving a coefficient of determination of 97.95%, thereby significantly improving forecasting accuracy and robustness.