Scholarlink

<p>This study aims to advance thermal management strategies for photovoltaic (PV) systems by evaluating the cooling efficiency of TiO<sub>2</sub>-water nanofluids and developing robust machine learning (ML) models for predicting surface temperature and power output. The primary objective is to enhance PV system performance and support intelligent decision-making in renewable energy applications. Experimental investigations were conducted using various nanofluid concentrations and flow rates, generating a dataset of 1000 observations. Seven ML algorithms were trained and assessed. The random forest (RF) model achieved the highest accuracy in predicting surface temperature, with an RMSE of 0.1491 and <i>R</i><sup>2</sup> of 0.9866. Using 0.1% TiO<sub>2</sub> nanofluid at 5 L&#xa0;min<sup>−1</sup> reduced PV surface temperature by 23.72% and improved power output by 16.36% compared to the uncooled system. In contrast, using water without nanoparticles led to a 2.94% increase in power. Visualization techniques, including violin plots and Taylor diagrams, confirmed the robustness of the RF model, revealing a median prediction error of 0.1&#xa0;°C, high <i>R</i><sup>2</sup> of 0.9933, and low standard deviation of 0.5731. The RF model maintained a mean relative residual of 0.38%. Other models, including XGBoost and artificial neural networks (ANN), also performed well, with <i>R</i><sup>2</sup> of 0.9835 and 0.9543, respectively. Support vector regression (SVR), while less accurate for temperature prediction (<i>R</i><sup>2</sup> = 0.7917), showed strong output power estimation (<i>R</i><sup>2</sup> = 0.9924). Bayesian Optimization further improved model accuracy, with XGBoost yielding the lowest MAE of 0.186. This study highlights the potential of integrating nanofluid-based cooling with data-driven tools in optimizing PV performance for sustainable energy systems.</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

Proactive thermal management of photovoltaic systems using nanofluid cooling and advanced machine learning models

  • 期刊论文

  • 发表日期：

    2025年10月3日

查看全文
  • Ismail Masalha,
  • Ali Alahmer,
  • Omar Badran,
  • Mustafa Awwad Al-Khawaldeh,
  • Siti Ujila Masuri,
  • Hussein Maaitah

摘要

This study aims to advance thermal management strategies for photovoltaic (PV) systems by evaluating the cooling efficiency of TiO2-water nanofluids and developing robust machine learning (ML) models for predicting surface temperature and power output. The primary objective is to enhance PV system performance and support intelligent decision-making in renewable energy applications. Experimental investigations were conducted using various nanofluid concentrations and flow rates, generating a dataset of 1000 observations. Seven ML algorithms were trained and assessed. The random forest (RF) model achieved the highest accuracy in predicting surface temperature, with an RMSE of 0.1491 and R2 of 0.9866. Using 0.1% TiO2 nanofluid at 5 L min−1 reduced PV surface temperature by 23.72% and improved power output by 16.36% compared to the uncooled system. In contrast, using water without nanoparticles led to a 2.94% increase in power. Visualization techniques, including violin plots and Taylor diagrams, confirmed the robustness of the RF model, revealing a median prediction error of 0.1 °C, high R2 of 0.9933, and low standard deviation of 0.5731. The RF model maintained a mean relative residual of 0.38%. Other models, including XGBoost and artificial neural networks (ANN), also performed well, with R2 of 0.9835 and 0.9543, respectively. Support vector regression (SVR), while less accurate for temperature prediction (R2 = 0.7917), showed strong output power estimation (R2 = 0.9924). Bayesian Optimization further improved model accuracy, with XGBoost yielding the lowest MAE of 0.186. This study highlights the potential of integrating nanofluid-based cooling with data-driven tools in optimizing PV performance for sustainable energy systems.