Lightweight hierarchical spatial feature extraction and sequential modeling for PV fault detection using pyramid network and GRU for edge applications

Solar photovoltaic (PV) systems are becoming an increasingly important source of renewable energy around the world. However, faults in these systems can drastically diminish energy production, resulting in economic losses and environmental issues. Traditional fault detection methods are based on manual examination, which can be time-consuming and labor-intensive. This study presents a Custom GRU Pyramid Network, a deep learning-based method for fault detection in solar PV systems. This uses a convolutional neural network (CNN) architecture to analyze images of solar PV panels and detect faults such as soiling, hotspots, and cracks. The proposed model integrates Spatial–Sequential modeling for feature refinement, leveraging pseudo-temporal GRU processing of spatial feature maps. The proposed model is trained using a dataset of Infrared solar module. The model's performance is measured using metrics such as accuracy, precision, and recall for 12 different classes. The proposed model is extremely light which is utilizing only 3.5 million parameters. The results reveal that the suggested GRU Custom Pyramid deep learning-based approach is highly accurate at detecting faults in solar PV systems. The model detects faults with 96% accuracy in 2-class and 91% in 12-class scenario, exceeding standard fault detection approaches. This technique can be integrated into existing solar PV monitoring systems, allowing for real-time fault identification and lower maintenance costs.