Advanced health-state data analytic workflow for utility-scale photovoltaic power plants

Abstract

This work aims to present data analytic advances and next-generation workflows for utility-scale photovoltaic (PV) power plant monitoring. The proposed health-state architecture comprises of an integrated and scalable workflow that includes data enrichment, predictive modelling and fault detection modules applied to high-resolution data streams. The obtained results demonstrated high power output predictive accuracies of <1.2%, given by the average root mean square error (RMSE) relative to the nominal capacity of the test-bench PV system, across different weather patterns and time durations. Furthermore, the robustness and location independency of the architecture was verified at utility-scale PV power plants by exhibiting high predictive accuracies. Moreover, the architecture proved capable to identify power, voltage and current failures with a detection accuracy of over 90%, even for low loss magnitudes. Finally, useful information is provided for establishing effective workflows for the performance evaluation of utility-scale power plants.