Practical implementation based on histogram of oriented gradient descriptor combined with deep learning: Towards intelligent monitoring of a photovoltaic power plant with robust faults predictions

The increasing complexity of photovoltaic (PV) system monitoring underscores the importance of precise fault detection and energy loss prediction. This paper proposes a deep learning-based framework that integrates multiple advanced techniques to accurately detect, localize, and predict faults in PV panels. A pre-trained Convolutional Neural Network (CNN), based on the AlexNet architecture, processes thermal imaging data for precise fault extraction. This facilitates the classification of faults, contributing to improved decision-making in PV system management.

To further enhance real-time monitoring, the framework integrates the Histogram of Oriented Gradients (HoG) descriptor with Support Vector Machine (SVM) models, enabling efficient detection and localization of hotspots across the panels. Additionally, the system leverages Long Short-Term Memory (LSTM) networks combined with fuzzy logic to predict panel performance degradation and quantify energy losses caused by detected faults. The learning process relies on the Long-Term Recurrent Convolutional Network (LRCN) to accurately forecast defects by analyzing power efficiency loss rates.

Experimental results confirm the effectiveness and reliability of the proposed framework. Achieving an accuracy of 95.45%, with a true positive rate of 91.67% and a true negative rate of 100%, the system demonstrates robust fault detection capabilities. These results highlight the framework’s potential to mitigate power losses, ensuring optimal operation of PV systems. This intelligent solution offers a significant advancement in PV system maintenance and monitoring, providing a scalable approach for real-world applications.