Research on Anomaly Detection and Root Cause Analysis of Performance Degradation in Photovoltaic Power Stations

Abstract

To address abnormal phenomena such as sudden power drops during photovoltaic power plant operation, this study proposes a data-driven anomaly detection and root cause analysis method to enhance system maintenance efficiency and power generation benefits. Utilising 5-minute operational data from a 50kW photovoltaic power plant, combined with NASA meteorological data, an integrated anomaly detection framework and multi-dimensional feature engineering system were constructed. K-Means++ clustering and LightGBM classification models were employed for pattern recognition and root cause tracing of detected anomalies. Experiments demonstrate: 1) The anomaly detection framework achieves an F1 score of 0.87; 2) Three distinct power loss patterns and their distribution ratios are identified; 3) The LightGBM root cause classification model attains an average accuracy of 88%. The proposed method effectively diagnoses the root causes of performance degradation in photovoltaic systems, providing both theoretical foundations and technical support for implementing predictive maintenance.