Abnormal Monitoring Method Of Wind Turbine Pitch System Based On Online Sequential Extreme Learning Machine

Under the influence of the current energy crisis and the global greenhouse effect,traditional thermal power generation which consumes fossil resources has causesd certain environmental pollution problems.Wind power generation with the characteristics of clean and renewable has attracted the attention of various countries in the field of new energy power generation.In recent years,China has been continuously increasing the scientific research and investment in wind power.With the increasing number of installed wind turbines in the world,the stablility and reliability of wind turbines,as well as the formulation of a safe and effective maintenance plan are particularly important,which requires anomaly monitoring and research on the key components of wind turbines.This thesis uses the data of the Supervisory Control And Data Acquisition(SCADA)of the wind turbine to realize the abnormal monitoring research of the wind turbine pitch system.This thesis focuses on the following parts:1)Based on mastering the basic structure and operating principles of doubly-fed wind turbines,the focus is on elaborating and analyzing the failure modes and causes of doubly-fed wind turbine pitch systems.Further,this thesis analyzes the characteristics of the monitored parameters and uses the ReliefF algorithm to realize the feature selection of the fan pitch system.2)Considering the complicated operating conditions of the pitch system,the strong nonlinearity among the monitoring variables,and the dynamic update of the data information of the SCADA system,a multi-parameter state monitoring model based on an online sequential limit learning machine was proposed.Aiming at the problem that the input weights and offsets of the online sequential extreme learning machine are generally random numbers,and the training effect of OS-ELM is greatly affected by the initial value,the quantum evolution algorithm is used to optimize the extreme parameter set of the extreme learning machine and improve the model Prediction accuracy.3)The multi-parameter state monitoring model of the pitch system based on the online sequential limit learning machine optimized by the quantum evolution algorithm was established.The residual set of the pitch system under the normal state was trained,and the residual signal of the model is calculated by using the Mahalanobis distance function.The anomaly detection threshold is determined by the Weibull distribution of Mahalanobis distance.when the actual Mahalanobis distance value obtained by the model exceeds the threshold,an anomaly alarm is generated.