基于相空间重构和门控循环单元的股价预测研究

准确预测股价波动是金融投资关注的焦点问题。股价波动受多种因素影响，具有非线性特征，传统的线性预测方法往往难以奏效。文章选择A股5个代表性股指与5只大市值股票为样本，使用其2020—2023年的日收盘价数据，首先，借助相空间重构技术（Phase Space Reconstruction, PSR）将股价时间序列映射到高维空间中，揭示其混沌特征；然后，基于门控循环单元（Gate Recurrent Unit, GRU）深度学习方法开发出PSR-GRU预测模型，生成股价预测结果；最后，将预测结果与经典预测模型所得结果进行对比。结果发现，股价波动具有混沌特性，PSR-GRU模型在股价预测上表现出更优异的性能。

It is critical to predict stock price fluctuations accurately in financial investment. Stock price fluctuations are influenced by multiple factors and have nonlinear characteristics, making traditional linear prediction methods often ineffective. We first selected 5 representative stock indices and 5 large-cap stocks of A-shares as samples, used their daily closing price data from 2020 to 2023, to map their price time series to a high-dimensional space by using the phase space reconstruction (PSR) technique, and reveal their chaotic characteristics. Then, based on the deep learning method Gated Recurrent Unit (GRU), we developed a PSR-GRU prediction method to generate stock price prediction results. Finally, we compared the predicted results with those obtained from classical prediction models. We found that stock price fluctuations have chaotic characteristics, and the PSR-GRU exhibits superior performance in stock price prediction.