Photovoltaics are a rapidly growing source of renewable energy. Wiring faults that plague photovoltaic systems, such as disconnects, ground faults, and transient arc faults, are difficult to diagnose and time-consuming to locate. Automatic fault detection in photovoltaic systems is thus desirable, but this is a challenge in the presence of benign environmental variables, such as temperature and illuminance, that dynamically distort data over time. Spread-spectrum time-domain reflectometry (SSTDR) is a promising fault detection method that has been proven capable of detecting ground and arc faults in photovoltaic systems as well as detecting and locating disconnects in strings of up to 26 panels. However, these experiments do not account for environmental change, and recent research has shown that variation due to environmental change is significant in SSTDR measurements. Current SSTDR fault detection compares measurements to a single baseline measurement, but measurements will deviate too far from the baseline in the presence of these environmental changes, obscuring the effects of faults.To improve the resilience of SSTDR to environmental variations, we create linear and mixed-effect regression models as well as feed-forward and recurrent neural networks that use illuminance, temperature, and humidity as input. These models are trained to predict the SSTDR measurements that would be expected from a fault-free string in a given environment. These predictions can be directly used as baseline measurements for fault detection. As a result, any deviation from these predicted baselines is more likely to be caused by faults, allowing for more confident and accurate fault diagnosis and localization. 
 
We train and test each model using an experimental dataset comprising three months of simultaneous SSTDR and environmental measurements. SSTDR measurements are taken from a string consisting of five solar panels, and the environmental variables measured were illuminance, temperature, and humidity. Models are evaluated and compared using the correlation coefficient between predicted baselines and their corresponding measured baselines. We find that recurrent neural networks produce the most accurate predictions, though all models produce encouraging results, with mean testing correlation coefficients between 0.958 (from the linear model) and 0.960 (from the recurrent neural network). This demonstrates how SSTDR fault detection can be made robust to long-term environmental change by predicting baselines from illuminance, temperature and humidity.

Baseline Signal Learning for Sstdr Fault Detection in Photovoltaic Strings

Paper Proceedings: https://asmedigitalcollection.asme.org/qnde/QNDE2021/volume/85529

ASME 2021 48th Annual Review of Progress in Quantitative Nondestructive Evaluation (QNDE2021)

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.