Trend-Based Predictive Maintenance and Fault Detection Analytics for Photovoltaic Power Plants

The generation of underperformance alerts is essential for ensuring the long-term performance and cost-effective operation and maintenance (O&M) of photovoltaic (PV) systems. Although various failure detection methods have been developed, a literature survey revealed a notable lack of strategies for effective predictive maintenance, that can accurately anticipate underperformance conditions. This study aims to develop an innovative workflow for fault detection and predictive maintenance in PV systems. The proposed workflow integrates a predictive model based on the eXtreme Gradient Boosting (XGBoost) algorithm to simulate the PV performance, the One-Class Support Vector Machine (SVM) algorithm for fault detection, and the Facebook Prophet algorithm to forecast performance trends and subsequently generate predictive maintenance alerts. The data-driven methodology was validated using a dataset from a 1.8 MWp PV power plant in Greece. The results showed the effectiveness of the workflow for fault detection and prediction. The workflow achieved a sensitivity of 96.9% for fault detection while when generating predictive maintenance alerts up to 7 days in advance, the workflow yielded a sensitivity of 92.9%.