Predictive Maintenance and Anomaly Detection Analytics for Utility-Scale Photovoltaic Plants

Ensuring optimal lifetime performance and reduced levelized cost of electricity (LCoE) of photovoltaic (PV) systems necessitates the anticipation of underperforming conditions and cost-effective maintenance interventions. Even though, many data-driven preventive maintenance techniques have been proposed, a main challenge remains the lack of a standardised solution for exploiting data stream signals to predict optimum operation and identify possible problems for utility-scale PV power plants. The scope of this work is to address this challenge by presenting a novel trend-based performance loss routine for predicting potential underperforming conditions at utility-scale PV power plants. The proposed predictive maintenance and anomaly detection workflow operates on acquired performance data streams and evaluates performance trend deviations by analysing performance deviations. The results demonstrated that the constructed workflow proved robust in predicting the performance (at both AC and DC sides) since absolute performance ratio (PR) deviations <0.3% were exhibited at different weather type categories. Furthermore, the trendbased evaluation routines were capable to predict potential induced degradation (PID) loss decreases once the difference between the predicted and forecasted daily average voltage slopes deviate >2% at utility-scale. Similarly, the predictive model for monitoring the insulation resistance (????????iso) demonstrated effectiveness to forecast low insulation incidences by analysing the daily ????????iso patterns. Finally, useful information is provided for the replication of predictive maintenance routines that form integrated components of utility-scale PV plant monitoring systems.