Scholarlink

<p>With the rapid development of Solar energy-based electricity generation, there is a need to know&#xa0;the fluctuation of solar irradiance in advance for the stability of the photovoltaic power system and safety of the power grid. Therefore, short-term irradiance predictions are widely used in decision support for power distribution systems. This paper contains the development of monthly forecast models based on deep learning sequence modelling for one-day ahead forecasts of half-hourly INSAT-3D Global Horizontal Irradiance (GHI) images. The architecture of the forecast model is decided by Network Architecture Search (NAS) algorithms by tuning hyperparameters in user provided search space. The developed month-wise models capture spatial–temporal relationships in historical GHI images and perform forecasting. The forecasting ability of the forecast models are evaluated using normalized Root Mean Square Error (nRMSE) and Structural Similarity Index Measure (SSIM) under monsoon and non-monsoon conditions. The models from the&#xa0;proposed methodology can predict images with nRMSE and SSIM of 8.08% and 91.84% respectively in non-monsoon months and with nRMSE and SSIM of 33.09 and 61.95% respectively in monsoon months over the&#xa0;Gujarat region of India.</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

One day Ahead Forecasting of INSAT-3D Solar Irradiance Using Automated Deep ConvLSTM Network

  • 期刊论文

  • 发表日期：

    2025年7月2日

查看全文
  • Ujjwal K. Gupta,
  • Shweta Mishra,
  • Shashikant A. Sharma

摘要

With the rapid development of Solar energy-based electricity generation, there is a need to know the fluctuation of solar irradiance in advance for the stability of the photovoltaic power system and safety of the power grid. Therefore, short-term irradiance predictions are widely used in decision support for power distribution systems. This paper contains the development of monthly forecast models based on deep learning sequence modelling for one-day ahead forecasts of half-hourly INSAT-3D Global Horizontal Irradiance (GHI) images. The architecture of the forecast model is decided by Network Architecture Search (NAS) algorithms by tuning hyperparameters in user provided search space. The developed month-wise models capture spatial–temporal relationships in historical GHI images and perform forecasting. The forecasting ability of the forecast models are evaluated using normalized Root Mean Square Error (nRMSE) and Structural Similarity Index Measure (SSIM) under monsoon and non-monsoon conditions. The models from the proposed methodology can predict images with nRMSE and SSIM of 8.08% and 91.84% respectively in non-monsoon months and with nRMSE and SSIM of 33.09 and 61.95% respectively in monsoon months over the Gujarat region of India.