Wind Turbine Condition Identification Method Based On SCADA Data

With the global emphasis on environmental protection issues and the rise of green energy,in order to solve the environmental pollution problems caused by traditional fossil energy sources,countries around the world have begun to vigorously promote the development of new energy industries,and the wind power industry as a high-quality green energy has therefore received more and more attention.With the continuous operation of wind farms,wind turbines inevitably age,which leads to frequent failures in the operation of wind turbines.The global downtime caused by wind turbine failure damage and maintenance has resulted in a large amount of wasted wind energy,which makes the wind power industry’s profits decline and is not conducive to industry development,so research related to wind power intelligent operation and maintenance has begun to receive attention.The current O&M strategy of traditional wind farm operators is mainly through regular inspection and maintenance activities and intervention after a fault has occurred.If we intervene after a fault has occurred,we can only react to the fault passively,and we cannot prevent the fault in advance and avoid the resulting damage.Periodic maintenance is not flexible enough and is prone to ineffective maintenance and excessive maintenance.Therefore,the idea of using the existing SCADA(Supervisory Control And Data Acquisition)system to confirm the status of the wind turbine and to provide early warning at the early signs of failure has started to receive attention from researchers.This idea not only solves the defects of traditional operation and maintenance methods,improves the efficiency of operation and maintenance,but also conforms to the development trend of wind power intelligence.In order to correctly detect the abnormal state of wind turbines,it is very important to pre-process the operating data of wind turbines to meet the data requirements of the established model.Through the study of wind speed and power data distribution characteristics,some of the data were screened out through statistical laws.For the distribution characteristics of the four types of common abnormal data of SCADA data,two clustering algorithms,KMeans and DBSCAN,were used to process and screen out the abnormal points.Subsequently,by studying the correlation of the parameters related to the inverter as the research object,a set of state parameters with high correlation with the operation status of the inverter was selected as the research object.This paper then proposes a similarity-based anomaly detection model to address the problem of misclassification of other anomaly detection models when problems occur with environmental parameter measurement sensors such as anemometers.Through the validation of wind speed data,the model is able to laterally use data from multiple similar turbines to find anomalies in anemometer data.The final chapter makes a study on the establishment and evaluation of the effect of the LSTM(Long Short Term Memory Neural Network)based wind turbine inverter condition detection model,discusses the working principle and modeling method of the model,and conducts experiments using the data input model under abnormal conditions to verify the feasibility of using the model for early warning of wind turbine faults.Subsequently,it is used in combination with the similarity-based wind speed anomaly detection model to demonstrate that the method can avoid model misclassification caused by wind speed data anomalies.