Scholarlink

<p>The F10.7 index serves as a key indicator of solar activity, and its accurate prediction is essential for mitigating adverse effects on radio communications, navigation systems, and satellite operations. This paper proposes a hybrid model integrating Variational Mode Decomposition (VMD), Second-Order Blind Identification (SOBI), and Deep Linear Network (DLinear) for medium-term prediction of the F10.7 index. The proposed approach begins by applying VMD to decompose the F10.7 time series into multiple modal components with distinct frequency characteristics, thereby reducing the inherent non-linearity and complexity of the data. Subsequently, SOBI is employed to extract independent source signals from these modal components, enhancing feature interpretability and discriminability. Finally, the DLinear model is utilized to model and predict the F10.7 index based on the refined features. Experimental results demonstrate that the VMD–SOBI–DLinear model achieves significant performance improvements in medium-term F10.7 forecasting. Compared to Informer, the proposed model exhibits enhanced prediction accuracy across the 1 – 27 day horizon, with MAE and RMSE reduced by 42.0% and 35.5%, respectively, at the 27-day prediction window. Moreover, the model maintains robust performance during both solar maximum and minimum periods, indicating its strong potential as a core tool for operational medium-term space weather 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

Medium-Term Forecasting of Solar F10.7 Using VMD-SOBI-DLinear Hybrid Model

  • 期刊论文

  • 发表日期：

    2026年4月27日

查看全文
  • Yang Guo,
  • Boyang Wang,
  • Xueshu Shi,
  • Jian Sun,
  • Jiaxing He

摘要

The F10.7 index serves as a key indicator of solar activity, and its accurate prediction is essential for mitigating adverse effects on radio communications, navigation systems, and satellite operations. This paper proposes a hybrid model integrating Variational Mode Decomposition (VMD), Second-Order Blind Identification (SOBI), and Deep Linear Network (DLinear) for medium-term prediction of the F10.7 index. The proposed approach begins by applying VMD to decompose the F10.7 time series into multiple modal components with distinct frequency characteristics, thereby reducing the inherent non-linearity and complexity of the data. Subsequently, SOBI is employed to extract independent source signals from these modal components, enhancing feature interpretability and discriminability. Finally, the DLinear model is utilized to model and predict the F10.7 index based on the refined features. Experimental results demonstrate that the VMD–SOBI–DLinear model achieves significant performance improvements in medium-term F10.7 forecasting. Compared to Informer, the proposed model exhibits enhanced prediction accuracy across the 1 – 27 day horizon, with MAE and RMSE reduced by 42.0% and 35.5%, respectively, at the 27-day prediction window. Moreover, the model maintains robust performance during both solar maximum and minimum periods, indicating its strong potential as a core tool for operational medium-term space weather forecasting.