Scholarlink

<p>Solar radiation is vital for producing renewable energy, requiring accurate forecasting to optimize photovoltaic systems and energy management. This research introduces a framework for Global Horizontal Irradiance (GHI) forecasting by combining Long Short-Term Memory networks (LSTM), Gated Recurrent Units (GRU) and Spatial–Temporal Attention (STA) mechanisms in a novel model referred to as GRU-STA-LSTM. This algorithmic framework introduces significant improvements in computational efficiency compared to existing models. Effectiveness is demonstrated through rigorous validation across diverse datasets, achieving higher accuracy and robustness under varying conditions. Utilizing a dataset of 33 years of hourly GHI data (1990–2023) from the Alturas and Meloland stations in California, the study systematically organizes the data into distinct lag period configurations. This approach captures complex temporal and spatial relationships, enhancing predictive accuracy. The model’s performance is assessed using various statistical metrics, including Mean Absolute Percentage Error, Normalized Root Mean Square Error, Mean Absolute Error, R-squared, Mean Bias Error, Nash–Sutcliffe Efficiency and the Willmott Index. The GRU-STA-LSTM model demonstrates MAPE of 4.5%, outperforming other models, including GRU-LSTM with a MAPE of 6.2% and LSTM with a MAPE of 6.7%. In addition, it achieves an R<sup>2</sup> value of 0.9636 and a significant reduction in MBE from 2.7568 W/m<sup>2</sup> at Seq 6 to 0.8626 W/m<sup>2</sup> at Seq 24. The NSE metric also reinforces its reliability, reaching 0.9636 at Seq 24. So, these results indicate that the GRU-STA-LSTM model captures the complexities of solar radiation patterns, thus validating its superiority over benchmark forecasting methods.</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

Innovative strategies for hourly global horizontal irradiance forecasting using a hybrid gated recurrent unit and long short-term memory architecture with spatio-temporal attention mechanism

  • 期刊论文

  • 发表日期：

    2025年6月18日

查看全文
  • Saeed Samadianfard,
  • Zahra Rousta,
  • Mahdi Mohebbiyan,
  • Hamed Talebi

摘要

Solar radiation is vital for producing renewable energy, requiring accurate forecasting to optimize photovoltaic systems and energy management. This research introduces a framework for Global Horizontal Irradiance (GHI) forecasting by combining Long Short-Term Memory networks (LSTM), Gated Recurrent Units (GRU) and Spatial–Temporal Attention (STA) mechanisms in a novel model referred to as GRU-STA-LSTM. This algorithmic framework introduces significant improvements in computational efficiency compared to existing models. Effectiveness is demonstrated through rigorous validation across diverse datasets, achieving higher accuracy and robustness under varying conditions. Utilizing a dataset of 33 years of hourly GHI data (1990–2023) from the Alturas and Meloland stations in California, the study systematically organizes the data into distinct lag period configurations. This approach captures complex temporal and spatial relationships, enhancing predictive accuracy. The model’s performance is assessed using various statistical metrics, including Mean Absolute Percentage Error, Normalized Root Mean Square Error, Mean Absolute Error, R-squared, Mean Bias Error, Nash–Sutcliffe Efficiency and the Willmott Index. The GRU-STA-LSTM model demonstrates MAPE of 4.5%, outperforming other models, including GRU-LSTM with a MAPE of 6.2% and LSTM with a MAPE of 6.7%. In addition, it achieves an R2 value of 0.9636 and a significant reduction in MBE from 2.7568 W/m2 at Seq 6 to 0.8626 W/m2 at Seq 24. The NSE metric also reinforces its reliability, reaching 0.9636 at Seq 24. So, these results indicate that the GRU-STA-LSTM model captures the complexities of solar radiation patterns, thus validating its superiority over benchmark forecasting methods.