State Detection Of Wind Turbine Based On Bidirectional Gated Recurrent Unit Neural Network

With the rapid development of science and technology,the problem of energy shortage and environmental pollution has become prominent.As a clean and renewable energy,wind energy has attracted attention from all over the world.In recent years,the wind power industry has developed rapidly,but with the increase of the operating time of wind power units,it will enter a period of high failure.The state detection and fault diagnosis of wind turbines,timely detection of early faults of wind turbines and the formulation of reasonable operation and maintenance plans are of great significance to ensure the long-term safe and reliable operation of wind turbines.Today,wind farms are equipped with Supervisory Control and Data Acquisition(SCADA),it records the wind turbines operating in the process of a large amount of data,if effective use of these data,extract the information related to the operation of wind turbines,used to detect the early abnormal state of wind turbines,the wind generator can effectively reduce the probability of major accidents,Improve the reliability of wind turbine operation and reduce the loss of wind farm operation and maintenance.In this paper,by mining the logical relationship between SCADA data and the operating state of wind turbines,the abnormal operating state of wind turbines is detected and studied.The main research contents are as follows:(1)Aiming at the problem that SCADA data of wind turbines have many variables and contain a large number of abnormal data,a state detection method of wind turbines based on CNN-GRU model is proposed.Abnormal data in SCADA data were eliminated according to quad analysis method,and Pearson correlation analysis method is used to analyze the correlation of the cleaned data,so as to determine the correlation between the variables,and select the variables with greater correlation,so as to reduce the dimension of the input data.Combined with the Convolutional Neural Network(CNN)and Gated Recurrent Unit(GRU),the CNN-GRU model was constructed to extract the spatial and temporal characteristics of SCADA.The normal state data after pretreatment was used as the input of the prediction model to train the normal state model.The SCADA data were input into the normal operating state model,and the variation trend of RMSE which predicted the residual was taken as the evaluation index.According to the normal state data,the threshold value was determined,and the early abnormal operating state of wind turbine was identified.(2)On the basis of CNN-GRU model,attention mechanism(AM)is added,and a wind turbine state detection method based on CNN-Bi GRU-AM model is proposed.Attention mechanism layer was added between CNN layer and Bi GRU layer to redistribute the weight of features,highlight the more important features and ignore the less important parts,and then optimize the model.Combined with the actual SCADA data,the normal operating state model was trained to predict the actual data,and the RMSE sequence of predicted residuals was calculated.The thresholds were determined based on the principle of exponential weighted moving average(EWMA),and the operating state of wind turbines was detected,and the early faults of wind turbines were identified.Compared with LSTM model,GRU model,BIGRU model and CNN-Bi GRU model,the results show that the CNN-Bi GRU model is more accurate and efficient in the detection of wind turbine operation state,and this method can more effectively improve the reliability and safety of wind turbine operation.