Accelerated and Accurate: A CNN-ResNet Based Fault Diagnosis Photovoltaic Systems

Photovoltaic (PV) systems are a backbone of the infrastructure of renewable energy with its usage growing significantly. Early fault detection of these systems being essential to enhance their reliability and efficiency. Despite the development of fault diagnosis methods of PV system promoted by machine learning models such as ensemble learning, support vector machine and neural networks, challenges in achieving high accuracy and generalization persist. In this paper, propose a deep learning method based on a ResNet architecture for reliable and efficient fault detection, including the following categories: Normal Operation, Short-Circuit, Degradation, Open Circuit, and Shadowing. Also devise a new learning rate schedule (LRS), which considerably improves the training dynamics and enables a ۷۴% improvement in model performance. The suggested method has excellent performance achieves ۹۹.۸% accuracy throughout the training, validation and testing phases. The results obtained showcase the potential of ResNet-based architectures, in addition to prowess in adaptive learning rate strategies, at enhancing the reliability of photovoltaic systems through scalable and precise fault diagnosis.

کلیدواژه ها:

Deep learning ، Fault detection ، Learning rate schedule ، Photovoltaic systems ، Renewable energy reliability