Scholarlink

<p>Accurate prediction of Global Horizontal Irradiance (GHI) is pivotal for optimizing solar energy generation, improving grid stability, and advancing climate forecasting. This study introduces an innovative hybrid deep learning framework, CNN-Informer-GRU, designed to enhance the precision and robustness of GHI forecasting. The model integrates the strengths of three advanced architectures such as CNN for extracting local features and detecting short-term fluctuations, the Informer model for capturing long-range temporal dependencies through efficient sparse attention mechanisms, and GRU for refining temporal dynamics to boost future predictions. To further elevate performance, a clustering technique such as DBSCAN is employed to group similar GHI patterns based on historical meteorological data. This multi-seasonal clustering spanning summer, rainy, and winter enables the model to focus on season-specific trends, thereby improving accuracy and generalization. The z-score normalization ensures data consistency and smooth convergence during training. The model's effectiveness is rigorously validated on real-world GHI datasets using statistical metrics such as MAE (3.111), MSE (26.8891), RMSE (5.1855), and NRMSE (0.0162), showcasing its superiority over traditional forecasting methods. This work highlights the critical role of hybrid deep learning and data-driven strategies in enabling reliable solar radiation forecasting an essential step toward sustainable energy systems and smarter energy infrastructure.</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

Accurate global horizontal irradiance estimation with hybrid convolutional neural network-informer-gated recurrent unit framework

  • 期刊论文

  • 发表日期：

    2026年3月16日

查看全文
  • Girijapati Sharma,
  • Rahul Gupta

摘要

Accurate prediction of Global Horizontal Irradiance (GHI) is pivotal for optimizing solar energy generation, improving grid stability, and advancing climate forecasting. This study introduces an innovative hybrid deep learning framework, CNN-Informer-GRU, designed to enhance the precision and robustness of GHI forecasting. The model integrates the strengths of three advanced architectures such as CNN for extracting local features and detecting short-term fluctuations, the Informer model for capturing long-range temporal dependencies through efficient sparse attention mechanisms, and GRU for refining temporal dynamics to boost future predictions. To further elevate performance, a clustering technique such as DBSCAN is employed to group similar GHI patterns based on historical meteorological data. This multi-seasonal clustering spanning summer, rainy, and winter enables the model to focus on season-specific trends, thereby improving accuracy and generalization. The z-score normalization ensures data consistency and smooth convergence during training. The model's effectiveness is rigorously validated on real-world GHI datasets using statistical metrics such as MAE (3.111), MSE (26.8891), RMSE (5.1855), and NRMSE (0.0162), showcasing its superiority over traditional forecasting methods. This work highlights the critical role of hybrid deep learning and data-driven strategies in enabling reliable solar radiation forecasting an essential step toward sustainable energy systems and smarter energy infrastructure.