Solar inverter event detection from four years of plant data
Four years of minute-level data from 30 inverters and 11 weather stations analysed: inverter events can be detected automatically with high confidence, and inverter output predicted from weather.
Challenge
A solar producer wanted to know whether the disturbances, warnings and grid events logged on its inverters could be detected, and ideally predicted, from the sensor and weather data it already collected.
Solution
Arundo processed about 250 GB of minute-level data from 30 inverters and 11 weather stations on a cluster-computing setup, cross-checked more than 36,000 historical events against the measurements, and tested multi-sensor anomaly detection for detection and pattern recognition for prediction, as well as weather-driven prediction of key inverter performance metrics.
Impact
Automatic event detection assessed as highly feasible; ahead-of-time prediction of the short, abrupt alarms assessed as unlikely with the data provided; inverter active power and daily energy shown predictable from weather. The findings fed a use-case workshop on inverter performance, predictive maintenance, alarm management and operational fingerprinting.