Parameter Optimization Based On SCADA Data Of Wind Turbine Pitch System Fault Early Warning

With the adjustment of the energy structure and the advancement of the energy revolution,it is of great significance to actively promote wind power generation to improve the energy structure and ease the energy crisis.However,since wind turbines are mostly located in remote areas with harsh envi ronments,the reliability and safety of wind power installations have become a recognized problem.Therefore,early warning of key components of wind turbines has far-reaching significance for the development of the wind power industry.In recent years,wi th the rise of machine learning and artificial intelligence,the combination of various intelligent algorithms and fault early warning methods has provided a new development direction for the field of fault early warning.Therefore,this paper applies the support vector regression algorithm to the wind turbine pitch system.In the fault early warning research,based on the SCADA operating data of the wind turbine,this paper proposes a fault warning for the wind turbine pitch system based on support vector regression.First,the research object of this paper is to select the doubly-fed variable-speed variable-speed asynchronous wind turbine,to understand the working principle of the pitch system and its operating status under different working conditions,a nd to analyze the typical faults and causes of the pitch system.Then,considering that most wind farms are equipped with SCADA systems,which record a large amount of wind turbine operating data,it can provide a wealth of information for wind turbine fau lt early warning research.According to the high-dimensional and non-linear characteristics of SCADA data,the data is pre-processed,and the fault feature vector of the pitch system is extracted using the Relief algorithm,and the rationality of the extra cted parameters is verified using the support vector machine classifier,which is useful for subsequent fault early warning.Modeling provides the foundation.Secondly,based on the fault feature vector of the pitch system extracted by the Relief algorithm,the support vector regression(SVR)algorithm is used to learn and train the behavior mode of the pitch system in normal operation,establish a fault early warning model of the pitch system,and select the grid search method.(GS)and genetic algorithm(GA)perform parameter optimization on SVR.The SCADA data of a wind farm in Zhangjiakou was used to train and test the model,and the model established by the results of the two optimization parameters was compared with the model established by the BP neur al network.The results show that the GA-SVR fault early warning model is accurate and optimal in model optimization.Both training speed and generalization have great advantages.Finally,considering the large changes in the operating conditions of wind turbines,the model parameter parameter output residuals fluctuate violently,and it is difficult to determine the time point for fault early warning.Therefore,a statistical process control chart(SPC)was introduced to calculate the residual fault early warning threshold and combined with information entropy The method quantifies the severity of the fluctuation of the residual error,and uses the two as a discriminant indicator of the fault early warning scheme.Finally,the SCADA data of a Zhangjiakou wi nd turbine unit running continuously before the shutdown on July 1,2018 was used as test sample data for an example simulation analysis to verify the feasibility of the proposed early warning scheme for the pitch system fault.