Research On Power And Health State Prediction Of Wind Turbines Based On PHM

In recent years,wind power has developed rapidly,but the grid connection of wind power is unstable,resulting in low wind power utilization.The construction of wind farms is remote and difficult to operate and maintain.At present,wind farms still adopt the methods of postinspection and periodic maintenance,and the maintenance costs during operation are relatively high.Precise forecasting of wind power can improve the stability of wind power grid integration and conduct more scientific grid dispatching,thereby effectively increasing the utilization rate of wind power.However,the accuracy of the current short-term power prediction model needs to be improved.If the decline trend and health status of the wind turbine can be accurately judged in advance through the monitoring data,the overhaul and maintenance of the equipment can be arranged reasonably,the quality of the normal operation of the unit can be improved,and the maintenance cost after the failure can be reduced,which has certain applicability and practical significance..Therefore,the research on predictive models is very necessary.This article takes the SCADA data of wind farms as the object.Due to the erroneous data in the historical wind power data,if the historical data is directly predicted,the prediction accuracy will be greatly affected.Therefore,this paper conducts research from three aspects: wind power historical data selection and reconstruction,short-term wind power prediction,and wind turbine health status prediction.The main work is as follows:(1)According to the analysis of the reasons for the wrong historical data of wind power,this paper adopts different screening methods and reconstruction methods for wind power historical data of different reasons.For the abandonment wind limit data,reconstruct it according to the power curve method;for the outlier data,use the quartile method to identify and reconstruct according to the interpolation method;and use the average relative error and accuracy as the basis for discrimination,and the quantitative analysis reiterates The accuracy of the structured data.(2)Aiming at the problem of low short-term prediction accuracy of wind power,this paper analyzes power data and proposes a short-term combined prediction model based on ISSA-RBF neural network and ARIMA.With the help of variational modal decomposition algorithm,the power data is decomposed into low-frequency and high-frequency components.For the low-frequency components,the ISSA-RBF model is used to predict them;for the high-frequency signal components,the ARIMA model,which is more applicable to nonstationary signals,is used for processing.Finally,the prediction results of each component are reconstructed,and the final prediction results are obtained.Further improve the prediction accuracy of the model.(3)Aiming at the problem of inaccurate judgment of the operating status of wind turbines,this paper establishes a wind turbine health prediction framework based on the real-time health index(RHI),and uses short-term combined forecasting models to obtain the standard residual set of wind turbine parameters and actual operating conditions.The real-time residual set.The Mahalanobis distance is used to describe the similarity of the two data sets,and the RHI value of the wind turbine is obtained after normalization.Case analysis shows that when the wind turbine operating state is abnormal,the prediction framework can predict the change trend of the health state of the wind turbine,which helps wind power plants to take maintenance measures in advance,reduce the frequency of turbine failures,and improve the reliability of operation.