Scholarlink

This study presents an innovative methodology for predicting solar irradiance and recalibrating weather stations using a hybrid model that combines gate recurrent units (GRU) and artificial neural networks (ANN). The model is trained on historical photovoltaic (PV) production data demonstrating exceptional accuracy with an RMSE of 29.5924, a MAPE of 6.68%, and an R2 of 0.9831. These results indicate that the model explains 98.31% of the variance in the data, validating its ability to predict solar irradiance with high reliability. Furthermore, after implementation, the model achieves an R2 of 0.96 when compared with pyranometer voltage measurements, confirming that the implementation produces highly reliable results. The model also enables the indirect recalibration of weather stations, using its predictions as a reference, thus avoiding the costs and effort associated with the physical recalibration of instruments. This approach contributes to optimizing photovoltaic systems and improving energy resource management, while supporting the transition to more sustainable renewable energy sources.

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

Optimization of Solar Irradiation Prediction and Recalibration of Weather Stations Using a Hybrid GRU-ANN Model

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mustapha Ezzini,
  • Raja Mouach,
  • Mohammed Ennejjar,
  • Abdelali El Gourari,
  • Mohammed Boukendil,
  • Mustapha Raoufi

摘要

This study presents an innovative methodology for predicting solar irradiance and recalibrating weather stations using a hybrid model that combines gate recurrent units (GRU) and artificial neural networks (ANN). The model is trained on historical photovoltaic (PV) production data demonstrating exceptional accuracy with an RMSE of 29.5924, a MAPE of 6.68%, and an R2 of 0.9831. These results indicate that the model explains 98.31% of the variance in the data, validating its ability to predict solar irradiance with high reliability. Furthermore, after implementation, the model achieves an R2 of 0.96 when compared with pyranometer voltage measurements, confirming that the implementation produces highly reliable results. The model also enables the indirect recalibration of weather stations, using its predictions as a reference, thus avoiding the costs and effort associated with the physical recalibration of instruments. This approach contributes to optimizing photovoltaic systems and improving energy resource management, while supporting the transition to more sustainable renewable energy sources.