统计与数据科学学院知识库

股票价格预测一直是金融领域中备受关注的话题，其研究意义深远而广泛，不管是对于投资者还是市场监管者而言均有着重要的现实意义。当前，随着数据科学、机器学习和人工智能的快速发展，对于股票变动趋势的预测正越来越精确以及多样化，股票市场研究正朝着更加复杂和多层次的方向发展。

本文所构建的VMD-CBAM-BiGRU模型为股票预测领域提供了一种兼具高精确性和稳定性的预测方法。首先，本文在基本行情指标的基础上引入大量的技术指标丰富特征集合，而后通过相关性分析筛选出对股价预测有显著效果的特征作为特征集合。在一系列基础模型中选择BiGRU作为基础预测模型，由于模型涉及大量超参数，合适的超参数可以显著提高模型的预测效果，而人工寻参效率低下，因此使用改进后的动态粒子群算法进行超参数寻优，提高模型的预测能力。由于股价序列具有高度非平稳性和强噪声以及大量特征等特点，考虑到VMD算法具有的序列降噪效果以及CBAM算法处理多特征数据中的优势，将VMD算法和CBAM算法引入BiGRU就基础模型进行降噪和特征提取处理，从两方面的优化中提高VMD-CBAM-BiGRU模型的预测效果。

本文从三个方面进行了模型的实证分析，首先在上证综指、沪深300、中国平安和贵州茅台四支股票上进行试验，验证了VMD算法和CBAM算法对于模型BiGRU显著的提升作用，证实了本文所构建VMD-CBAM-BiGRU模型在股票预测领域的稳定性和有效性，然后进行跨时期股价预测验证了模型在中短期预测中的可行性和有效性，最后使用所构建的模型作为交易策略进行模拟交易验证了模型在投资应用中的实用效果，从而为股市投资者和监管部门预测股价动向提供了一种优秀的预测工具。

Stock price prediction has always been a topic of concern in the financial field, and its research significance is far-reaching and extensive, which has important practical significance for both investors and market regulators. At present, with the rapid development of data science, machine learning and artificial intelligence, Forecasts of stock movements are becoming more accurate and diversified ,stock market research is moving towards a more complex and multi-layered direction.

VMD-CBAM-BiGRU model is constructed to provide a method with high accuracy and stability for stock forecasting. First of all, a large number of rich feature sets of technical indicators are introduced on the basis of basic market indicators, and then selects the features that have significant effect on stock price prediction as the feature set through correlation analysis. BiGRU is selected as the basic prediction model among a series of basic models. Since the model involves a large number of hyperparameters, Suitable hyperparameters can significantly improve the prediction effect of the model, while manual parameter seeking is inefficient, particle swarm optimization will be used to optimize the hyperparameters and improve the prediction ability of the model. Since stock price sequence has high non-stationary, strong noise and a large number of features, VMD algorithm and CBAM algorithm are introduced into BiGRU to carry out noise reduction and feature extraction processing on the basic model, and improve the prediction effect of VMD-CBAM-BIGRU model from the two aspects of optimization.

Empirical analysis of the model are conducted from three aspects. First, experiments were conducted on four stocks, namely Shanghai Composite Index, Husen-Shenzhen 300, Ping An of China and Kweichow Moutai, to verify that VMD algorithm and CBAM algorithm significantly improved BiGRU model. The stability and effectiveness of the VMD-CBAM-BiGRU model in the field of stock prediction was verified. Then, the feasibility and effectiveness of the model in the short and medium term prediction was verified by inter-period stock price prediction. Finally, the model was used as a trading strategy for simulated trading to verify the practical effect of the model in investment application. Thus, it provides an excellent forecasting tool for stock market investors and regulators to predict the trend of stock prices.