Research On Fault Diagnosis Of Wind Turbine Variable Pitch System

The complexity structure,the harsh operating conditions and the diversity of related factors of the wind turbine cause the failure rate increases continually.The variable pitch system is one of the core control technology of wind turbine.How to reduce the failure rate of variable pitch system,the down time and the operation cost as well as improve the efficiency of the operation and maintenance of the wind turbine is a problem,which has to be solved as soon as possible for the investment,construction,operation and maintenance of wind power.By introducing the basic structure and working principle of wind turbine variable pitch system,this thesis analyzes the variable pitch system failure modes,which lay foundation for the fault diagnosis.With the running state of the wind turbine changing from normal to fault,the running data changes gradually.Based on the Supervisory Control and Data Acquisition(SCADA)systems’ data of 1.5MW running grid-connected wind turbines of some wind farm in Hebei Province,the operating data of the normal state and three common kinds of variable pitch fault state are collected.By using the proposed method of data preprocessing and pitch fault diagnosis,not only the feature information of data can be excavated,but also the fault diagnosis and fault classification of the variable pitch system can be carried out.In view of the instability of the operating condition,the complex failure reasons and strong nonlinear in parameters,Particle Swarm Optimization for Least Square Support Vector Machine(PSO-LSSVM)is proposed as a fault diagnosis method.Least Square Support Vector Machine(LSSVM)is very suitable for wind turbine variable pitch system fault diagnosis modeling,because it adopts the principle of minimized structural risk to replace the traditional principle of minimized empirical risk,which has strong learning ability,generalization ability and obvious advantages in solving small sample,nonlinear and high dimensional problems.In order to reach the highest accuracy requirements of fault diagnosis,Particle Swarm Optimization(PSO)is used to optimize the parameters of the radial basis kernel function LSSVM,namely: penalty factor C and kernel width parameter ?.The PSO-LSSVM is used to construct the pitch angle fault diagnosis model,the motor speed fault diagnosis model and the power output fault diagnosis model.The accuracy and effectiveness of the models have been demonstrated using the root mean square error and the diagnostic accuracy.The results of the simulation have demonstrated that the proposed approach has strong protential for wind turbine pitch system fault diagnosis.