Fault Prediction Model of Wind Power Pitch System Based on BP Neural Network

Abstract

Pitch system fault prediction and improvement of prediction accuracy are key technologies for wind power devel-opment, which ensure safe operation of the grid and effectively reduces operation and maintenance costs. The Supervisory con-trol and data acquisition (SCADA) system data is analyzed and processed to extract the associated parameters, i.e. output power, wind speed, pitch angle, and rotor speed. A Back Propagation (BP) neural network is used to train the system, taking into account the volatility and uncertainty of wind turbine param-eters, and a regression prediction model with a support vector regression (SVR) algorithm is also used for training. A pitch failure prediction model is established to predict the operation of the pitch system, which is used to develop a reasonable operation and maintenance plan. Through the system simulation, the prediction model performance index, error-index, and output data graphics are compared and analyzed.