Contactless Diagnosis Method and Unsupervised Learning for Panel-Level Photovoltaic Plant Operation and Maintenance

Abstract

Renewable energy systems that depend on photovoltaic energy are crucial in the shift towards sustainable energy. Detecting faults in photovoltaic components is highly valued. However, most research on fault detection relies on detailed electricity data for each panel of the photovoltaic station. In contrast, our study introduces a panel-contact-free algorithm for diagnosing faults using only whole station parameters. This algorithm enables the overall detection of panel faults, reducing the number of necessary sensors to only the irradiance detector while maintaining high accuracy in fault locating and classifying. Using fewer sensors minimizes the impact of power plant modifications and reduces overall transformation costs. Our study, which used actual PV power plant data for modeling, achieved a specific fault diagnosis accuracy of 93.18%, indicating strong potential for practical application and value.

Keywords

• Contactless diagnostics
• deep learning
• photovoltaic
• unsupervised learning