Automatic IV Curve Diagnosis with Deep Learning

Abstract

IV curve tracing is a useful method for diagnosing PV array underperformance problems. Its shape and values can reveal internal health issues of devices, caused by degradation, mismatch, cell cracks, or external issues of operating environment, such as shading and soiling. In recent years, there is a trend for string inverters to provide IV scanning function, which enables large scale high throughput IV curve diagnosis for PV farms. For this application, it is important to have automatic diagnosis and reporting instead of manual interpretation. In this work, we propose a diagnosis framework based on deep learning to classify various DC underperformance issues or faults from IV curves. The initial model training is accomplished by simulation of IV curves for a wide range of possible scenarios. Preliminary results indicate that the model is capable of discerning major classes of issues to a very high degree of accuracy.