Scholarlink

<p>The growing size of photovoltaic (PV) power systems frequently requires certain timely guaranteed fault diagnosis with an utmost precise method to make sure of maximum power production and minimal maintenance expense. The conventional parameter calculations primarily focus on electrical parameters, usually neglecting local thermal irregularities, which are indications of panel degradation. This paper presents a hybrid AI system consisting of sensor data classification with Extreme Gradient Boosting (XGBoost) and another thermal image analysis with Convolutional Neural Networks (CNN). The multi-modal dataset involves electrical couples’ features (voltage, current, power, temperature, irradiance, and humidity) together with infrared thermal images for arc fault detection, hot spots detection, and shading effects detection. Wavelet-based feature extraction techniques can enhance XGBoost, CNN reduces unsupervised heat-related anomalies drastically, and their weighted fusion is at a staggering 99.42% fault detection rate, outperforming plain machine learning or deep learning classifiers. The lightweight TensorFlow Lite deployment guarantees real-time operation on edge devices and further enables autonomous inspections through drones or IoT monitoring devices. Through experimental verification, the framework maintains robustness, scalability, and interpretability and is commercially viable for extremely large solar farms. By combining thermal and electrical concerns in a single platform, rather elegant design can be used in predictive maintenance methods to reduce false alarms and improve operational reliability-the major concerns of renewable energy monitoring-working towards the more sustainable process of solar power generation.</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

Hybrid AI Framework for Photovoltaic Fault Detection Using XGBoost and CNN with Real-Time Edge Deployment

  • 期刊论文

  • 发表日期：

    2026年2月20日

查看全文
  • T. Tamil Selvi,
  • R. Ramaprabha

摘要

The growing size of photovoltaic (PV) power systems frequently requires certain timely guaranteed fault diagnosis with an utmost precise method to make sure of maximum power production and minimal maintenance expense. The conventional parameter calculations primarily focus on electrical parameters, usually neglecting local thermal irregularities, which are indications of panel degradation. This paper presents a hybrid AI system consisting of sensor data classification with Extreme Gradient Boosting (XGBoost) and another thermal image analysis with Convolutional Neural Networks (CNN). The multi-modal dataset involves electrical couples’ features (voltage, current, power, temperature, irradiance, and humidity) together with infrared thermal images for arc fault detection, hot spots detection, and shading effects detection. Wavelet-based feature extraction techniques can enhance XGBoost, CNN reduces unsupervised heat-related anomalies drastically, and their weighted fusion is at a staggering 99.42% fault detection rate, outperforming plain machine learning or deep learning classifiers. The lightweight TensorFlow Lite deployment guarantees real-time operation on edge devices and further enables autonomous inspections through drones or IoT monitoring devices. Through experimental verification, the framework maintains robustness, scalability, and interpretability and is commercially viable for extremely large solar farms. By combining thermal and electrical concerns in a single platform, rather elegant design can be used in predictive maintenance methods to reduce false alarms and improve operational reliability-the major concerns of renewable energy monitoring-working towards the more sustainable process of solar power generation.