Scholarlink

<p>In the era of renewable energy, a precise and accurate solar forecasting is essential for efficient power system. This paper presents a systematic review of Deep Learning Neural Network (DLNN) models for PV power output and solar irradiance forecasting. The study analyses data preprocessing techniques like signal decompositions and outlier detection. It also explores the conventional to advanced features extraction strategies for DLNN. The ability of DLNN models to capture the complex temporal and spatial dynamics feature are also discussed. An especial emphasis is given on hybrid models that combine stacking algorithms, signal decomposition, and intelligent optimization techniques to improve predictive accuracy. Moreover, recent advancements in model fine-tuning using bio-inspired optimization algorithms are also discussed. By insights from hundreds of studies, this review highlights key challenges, limited real-world deployment of DLNN based systems and the underutilization of advanced bidirectional and attention-based models. Finally, the paper outlines promising avenues for future research, including the incorporation of multi-modal feature integration, and interpretable machine learning models. These insights aim to help the establishment of more robust, scalable, and accurate solar forecasting models, critical to the global transition toward sustainable energy.</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

State-of-the-art deep learning models for solar forecasting: models, methodologies, and key challenges

  • 期刊论文

  • 发表日期：

    2026年7月13日

查看全文
  • Rakesh Kumar Mittal,
  • Rashmi Gupta,
  • Anil Kumar Mishra,
  • Pardeep Singla

摘要

In the era of renewable energy, a precise and accurate solar forecasting is essential for efficient power system. This paper presents a systematic review of Deep Learning Neural Network (DLNN) models for PV power output and solar irradiance forecasting. The study analyses data preprocessing techniques like signal decompositions and outlier detection. It also explores the conventional to advanced features extraction strategies for DLNN. The ability of DLNN models to capture the complex temporal and spatial dynamics feature are also discussed. An especial emphasis is given on hybrid models that combine stacking algorithms, signal decomposition, and intelligent optimization techniques to improve predictive accuracy. Moreover, recent advancements in model fine-tuning using bio-inspired optimization algorithms are also discussed. By insights from hundreds of studies, this review highlights key challenges, limited real-world deployment of DLNN based systems and the underutilization of advanced bidirectional and attention-based models. Finally, the paper outlines promising avenues for future research, including the incorporation of multi-modal feature integration, and interpretable machine learning models. These insights aim to help the establishment of more robust, scalable, and accurate solar forecasting models, critical to the global transition toward sustainable energy.