Wind Turbine Power-Curve Modelling and SCADA Anomaly Detection

Wind-farm SCADA systems log wind speed, direction, power, pitch and temperatures every 10 minutes. A normal-behaviour model of the power curve reveals under-performance - stops, icing, pitch faults, derating - as soon as the measured power falls below the expected power. This project builds, for each of the six Senvion MM92 turbines (2.05 MW) of the Kelmarsh wind farm in the UK, the classical IEC 61400-12-1 binned power curve with air-density correction and compares it with k-nearest-neighbour, LightGBM gradient-boosting and MLP neural-network models that use the wind speed, air density, direction, the neighbour turbines' wind speed and the pitch angle. The residual of the deployed model drives EWMA and CUSUM alarms whose limits come from clean training data; the alarms are checked against the turbines' logged events, a fault-injection study and the operator's lost-energy figures. Fit on 2017, tested once on 2018. A web application with five databases, a GNU Octave cross-check (run) and a complete MATLAB version (written, not run) are included. Everything runs on a student laptop CPU in about 15 minutes.

Category: Electrical and Electronics Engineering (Renewable Energy, Wind Power, Condition Monitoring, Power-Performance Testing IEC 61400-12-1, SCADA Analytics, Machine Learning, MATLAB, GNU Octave, Python)

Electrical & Electronics Engineering Project EE03 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.