Scholarlink

<p>This study addresses the need for improved Air Temperature (AT) forecasting during extreme thermal conditions, as it is crucial due to the major influence such conditions exert on various activities. However, an integrated hybrid model is proposed, leveraging the capabilities of neural networks to enhance predictive accuracy and Machine Learning (ML) techniques to assess Photovoltaic (PV) installation risks, using the Power Temperature Coefficient for vigilance during extreme thermal events. Neural networks are designed to capture complex patterns and dependencies in the data, enabling accurate forecasts of upcoming AT fluctuations; while, ML techniques are adapted to determine the specific hotspot where the photovoltaic plant is situated. The obtained results indicate that the Air Temperature Forecaster through Recursive Model-based Convolutional Neural Network (ATFRM-CNN) consistently excels in medium- to long-term predictions (36–72&#xa0;h), achieving high correlation coefficient (r) values (<i>r</i> = 97.29–94.90%) and a low mean absolute error of 1.168&#xa0;°C for forecasts up to 72&#xa0;h. The ATFRM-based Recurrent Neural Network (ATFRM-RNN) and ATFRM-based Gated Recurrent Unit (ATFRM-GRU) models perform best for short-term predictions (6&#xa0;h) but exhibit a significant decline in accuracy over longer prediction horizons. Additionally, the results of classification techniques in assessing the risks associated with PV plants demonstrate that Gradient Boosting and Random Forest models exhibit exceptional performance, achieving 98% accuracy, surpassing the other classification models. In contrast, Support Vector Machines underperform with an accuracy of 52%.</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

Enhancing machine learning model for early warning in PV plants: air temperature prediction informed by power temperature coefficient

  • 期刊论文

  • 发表日期：

    2025年1月14日

查看全文
  • Mohamed Khala,
  • Naima El Yanboiy,
  • Ismail Elabbassi,
  • Omar Eloutassi,
  • Mohammed Halimi,
  • Youssef El Hassouani,
  • Choukri Messaoudi

摘要

This study addresses the need for improved Air Temperature (AT) forecasting during extreme thermal conditions, as it is crucial due to the major influence such conditions exert on various activities. However, an integrated hybrid model is proposed, leveraging the capabilities of neural networks to enhance predictive accuracy and Machine Learning (ML) techniques to assess Photovoltaic (PV) installation risks, using the Power Temperature Coefficient for vigilance during extreme thermal events. Neural networks are designed to capture complex patterns and dependencies in the data, enabling accurate forecasts of upcoming AT fluctuations; while, ML techniques are adapted to determine the specific hotspot where the photovoltaic plant is situated. The obtained results indicate that the Air Temperature Forecaster through Recursive Model-based Convolutional Neural Network (ATFRM-CNN) consistently excels in medium- to long-term predictions (36–72 h), achieving high correlation coefficient (r) values (r = 97.29–94.90%) and a low mean absolute error of 1.168 °C for forecasts up to 72 h. The ATFRM-based Recurrent Neural Network (ATFRM-RNN) and ATFRM-based Gated Recurrent Unit (ATFRM-GRU) models perform best for short-term predictions (6 h) but exhibit a significant decline in accuracy over longer prediction horizons. Additionally, the results of classification techniques in assessing the risks associated with PV plants demonstrate that Gradient Boosting and Random Forest models exhibit exceptional performance, achieving 98% accuracy, surpassing the other classification models. In contrast, Support Vector Machines underperform with an accuracy of 52%.