Detecting Anomalous Events for a Grid Connected PV Power Plant Using Sensor Data

This paper presents an anomaly detection algorithm for a large scale grid-connected photovoltaic (PV) power plant. Expected string current is calculated from measured weather data and compared to the measured current. The Python PV systems modelling library, `PVLib', is used for modelling the expected current of the plant for a given location. The detection algorithm is able to successfully flag anomalous events. Using the algorithm on the testing dataset, ranging from 2015 to 2018, 7516 events are flagged as having unexpected behaviour. A plant performance map is developed and shows potential in verifying flagged events. The real-time use of the performance map might also hold potential for identifying under-performing regions in the power plant.