A Data-Driven PCA–OCSVM Framework for Intelligent Monitoring and Anomaly Detection of Grid-Connected PV Inverters Under Multitask Operation

This study proposes an unsupervised anomaly detection method to identify the performance degradation in grid-connected photovoltaic (PV) inverters under multitask operation. Principal Component Analysis (PCA) and One-Class Support Vector Machine (OCSVM) were integrated to build a detection model using routine operational data. The key features include DC input, AC output, AC/DC ratio, and AC power variation, which are reduced to two principal components for anomaly boundary construction. The inverters were flagged as degraded if the AC/DC ratio was <0.96, the power fluctuation exceeded 20%, or the data fell outside the OCSVM-defined boundary. Compared with the Isolation Forest, the proposed method showed higher sensitivity. When applied to a 120 MW PV plant in Taiwan with 1292 inverters, including 55 PV-STATCOM units at night, the framework detected degradation in 5.4% of them. These results support their use in intelligent monitoring and predictive maintenance. In addition, through early fault detection and maintenance prioritization, the proposed framework contributes to enhancing reliability, reducing maintenance costs, and promoting the sustainable operation of utility-scale photovoltaic power plants.