Research On Fault Diagonosis Method Of Wind Turbine Blade Based On Feature Mining Of Time Series Data

Wind energy,as a kind of green and renewable energy,has gradually attracted worldwide attention.Due to wind turbines is located in the remote and harsh,higher failure rate of the unit,and has been found that there are many factors cause the malfunction of the wind turbine so if it can be found from the wind turbine itself on the run as soon as possible in the process of internal information,can according to have mastered the basic information of the wind turbine corresponding safeguard measures as soon as possible,to ensure the safety of wind turbine continues to run effectively,It is of great significance to prolong the whole life cycle of the wind turbine.In the current situation of rapid development of industrial big data,how to make reasonable and efficient use of the large amount of data generated during the operation of the wind turbine,and conduct in-depth mining and exploration of the data,to find the internal associated information generated during the operation of the wind turbine.Therefore,it has become a hot topic in the field of wind power fault diagnosis to find an appropriate fault diagnosis method for the whole wind turbine and its components and build the corresponding model to guarantee the healthy operation of the wind turbine.This paper takes the wind turbine blade as the research object and builds the corresponding fault diagnosis model for the phenomenon of blade icing.Based on the pre-installed Data of Supervisory Control and Data Acquisition(SCADA)system of wind turbine,the fault diagnosis model of wind turbine blade icing is established.Through in-depth analysis of the relationship between blade health state and SCADA data,combined with the time series characteristics of SCADA data,a deep learning based fault diagnosis method for wind turbine blades is studied,aiming at achieving high-precision classification diagnosis of wind turbine blades.The main work of this paper is as follows:(1)The common fault types and fault causes of wind turbine blades were systematically sorted out,and the basic principle of blade fault diagnosis based on SCADA data was analyzed in detail.In addition,the relationship between SCADA data and blade icing state was discussed and analyzed with the ice fault of wind turbine blades as the research object,which laid a foundation for subsequent model research.(2)A Wavelet-based Multiscale Long Short-Term Memory(Wavelet LSTM)model for leaf ice diagnosis was proposed based on SCADA data with time-series non-stability and multi-scale characteristics.Firstly,the original signal is decomposed by discrete wavelet according to the multi-scale characteristics of wavelet,and the local signals at different scales are obtained.Then,in the time feature learning stage,a fixed LSTM layer was used to extract the features of each decomposition level at the local scale and the features of original data at the global scale,respectively,and the time feature learning at different scales was independent of each other.Finally,the extracted features of all scales are stitched together,and the final detection classification is realized through the full connection layer and the Softmax layer.Based on the actual data collected from a wind field,the test of blade icing anomaly detection and diagnosis was carried out to verify the effectiveness of the proposed method.(3)Considering the high dimensional redundancy of SCADA data and the strong spatial correlation between different variables,a Weighted Multiscale Convolutional Neural Network(WMCNN)based blade icing fault diagnosis method was proposed.Firstly,considering that the operating state of wind turbine has multiple operating conditions,the influence weight of each dimension feature on the final result is different under different operating conditions.By introducing the attention mechanism,the weight of each dimension feature is dynamically changed according to different operating conditions.Secondly,SCADA data collected by various sensors of the system also have spatial correlation,and convolutional neural network is used to mine the spatial correlation among data information from spatial dimension.The abnormal detection of leaf icing was realized by training the model with labeled leaf icing data.Finally,the effectiveness of the proposed method compared with traditional methods and other deep learning methods is verified by using the actual wind field data.