Scholarlink

<p>In the realm of financial markets, accurately predicting stock prices presents both a significant research challenge and a practical necessity for investors and financial institutions. While traditional methods based on economic and financial theories provide a foundational framework, their limitations in capturing non-linear and dynamic patterns necessitate the exploration of advanced predictive techniques. This study addresses the challenges associated with financial time series prediction by proposing the CEEMDAN-GRU-Informer model within a deep learning framework. By incorporating the latest Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) technology, the original financial time series is decomposed to extract price variation information across different time scales. The complexity of each decomposed sequence is assessed using the Sample Entropy (SE) method, facilitating the selection of an appropriate predictive model. In the model prediction phase, a hybrid approach combining Gated Recurrent Unit (GRU) and Informer architectures is employed to enhance forecasting accuracy, leveraging various data-driven techniques to adapt to the complexity of the sequences. This research aims to significantly improve forecasting capabilities, offering stakeholders valuable insights for decision-making in volatile and dynamic financial environments. Empirical studies demonstrate that, within the SZ1 and SZ6 datasets, the root mean square error (RMSE) of the proposed model is 0.0007 and 0.0009, respectively, indicating substantial improvements over other hybrid models and highlighting its suitability for financial time series forecasting.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Application of the Deep Learning Integrated Framework CEEMDAN-GRU-Informer in Financial Time Series Prediction

  • 期刊论文

  • 发表日期：

    2025年6月18日

查看全文
  • Fang Liu,
  • Chen Liang,
  • Zhihan Jia

摘要

In the realm of financial markets, accurately predicting stock prices presents both a significant research challenge and a practical necessity for investors and financial institutions. While traditional methods based on economic and financial theories provide a foundational framework, their limitations in capturing non-linear and dynamic patterns necessitate the exploration of advanced predictive techniques. This study addresses the challenges associated with financial time series prediction by proposing the CEEMDAN-GRU-Informer model within a deep learning framework. By incorporating the latest Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) technology, the original financial time series is decomposed to extract price variation information across different time scales. The complexity of each decomposed sequence is assessed using the Sample Entropy (SE) method, facilitating the selection of an appropriate predictive model. In the model prediction phase, a hybrid approach combining Gated Recurrent Unit (GRU) and Informer architectures is employed to enhance forecasting accuracy, leveraging various data-driven techniques to adapt to the complexity of the sequences. This research aims to significantly improve forecasting capabilities, offering stakeholders valuable insights for decision-making in volatile and dynamic financial environments. Empirical studies demonstrate that, within the SZ1 and SZ6 datasets, the root mean square error (RMSE) of the proposed model is 0.0007 and 0.0009, respectively, indicating substantial improvements over other hybrid models and highlighting its suitability for financial time series forecasting.