基于注意力机制融合TCN和LSTM的股价预测方法——以中国银行(601988)为例

Stock Price Prediction Based on Attention-Fused TCN and LSTM—Taking Bank of China (601988) as an Example

摘要: 随着我国金融市场的不断深化与发展，股票价格的波动预测成为投资者和研究者关注的焦点。本文以中国银行(601988)为研究对象，使用Python爬取英为财经网站上的股价数据。在此基础上，提出了一种基于注意力机制融合时间卷积网络(TCN)和长短期记忆网络(LSTM)的股价预测模型。通过对中国银行股票的开盘价格进行训练与预测，本文所提出的模型在预测精度上优于传统的ARIMA模型和单一的LSTM模型。实验结果显示，注意力机制融合TCN + LSTM模型在预测误差和损失函数MSE上均表现出较好的性能，尤其在处理股票价格的极值点和长期趋势方面具有显著优势。本研究不仅为股票投资者提供了有效的预测工具，也为金融时间序列分析领域提供了新的研究视角。

Abstract: As China’s financial market continues to deepen and evolve, forecasting stock price fluctuations has become a focal point for both investors and researchers. In this study, China Bank (601988) is selected as the case study, and stock price data is collected using Python from the Investing.com website. Building on this foundation, we propose a novel stock price prediction model that integrates features from a Temporal Convolutional Network (TCN) and a Long Short-Term Memory (LSTM) network through an Attention Mechanism. By training and forecasting the opening prices of China Bank’s stocks, our results demonstrate that the proposed model achieves higher prediction accuracy than both the traditional ARIMA model and the standalone LSTM model. Experimental outcomes further indicate that the attention-fused TCN + LSTM model outperforms in terms of prediction error and Mean Squared Error (MSE), particularly excelling in capturing extreme price fluctuations and long-term trends. This study not only provides a valuable predictive tool for investors but also offers a fresh perspective for financial time series analysis.