Wind Turbine Fault Detection Method Based On Deep Fusion Of SCADA Data Features

With the continuous development of the wind power industry,the capacity of wind turbines is increasing,and reducing maintenance costs has become an urgent problem to be solved.The condition monitoring and fault diagnosis technology of wind turbine has become the key technical means to solve these problems.Based on the measured data of the wind turbine data acquisition and data acquisition(SCADA)system,the paper addresses the low accuracy of wind turbine fault prediction and the defects of traditional fault prediction methods,and combines Convolutional Neural Network(CNN),Long-term and short-term memory networks(LSTM),Self-attentive Mechanism(SM)and other related techniques to mine the wind turbine fault information hidden in SCADA data.The main tasks are as follows.(1)To address the complex parameters of wind turbine SCAD A data and the problem of inaccurate fault prediction due to local changes in data due to climate and other factors,a wind turbine fault prediction method with improved Tre Net model is proposed.The method improves the Tre Net model based on the Inception structure,and achieves multi-scale spatial features through different convolutional kernel size convolutional neural networks to capture the local variation information of fault features;long and short-term memory networks for global dependency extraction of fault features;and feature fusion of local information and global dependency of fault features through feature fusion layer.Through the field data verification,the method achieves the early identification of wind turbine gearbox faults.(2)To address the problems of low fault prediction accuracy and false alarm due to the coarse-grained characteristics of data brought by the sampling mode of SCADA data,as well as the volatility and uncertainty of monitoring information affected by random changes of wind speed and wind direction.Combining the self-attentive mechanism,cavity convolutional neural network(ACNN)and bi-directional long and short-term memory network(Bi LSTM),we propose a wind turbine prediction method based on the self-attentive mechanism of spatio-temporal cascade depth fusion.The method performs spatial feature extraction between different components by cavity convolution;the bi-directional long and short-term memory network realizes time-dependent extraction and spatio-temporal feature fusion in spatial features;and the fault features are weighted by multi-layer self-attentive mechanism layers.After field data verification,the method effectively eliminates false alarms and achieves accurate prediction of wind turbine faults.