Wind Turbines Diagnosis And Early Warning Based On Deep Learning

Due to operate in bad environment for long-term,wind turbines failures occur frequently.However,when using the traditional signal method,due to the strong noise problem of wind power signal itself and the limitation of monitoring signal fluctuation,the effect is not satisfactory.Therefore,this paper studies the wind turbine fault diagnosis and early warning method based on deep learning,aiming to apply the rapid and mature deep learning in recent years to maintain the safe and stable operation of wind turbines.Firstly,according to the characteristics of wind turbine data,an intelligent fault diagnosis method based on improved DenseNet is designed.This method develops a new coding method for the wind turbine data,inputs the historical data collected by the wind power plant data acquisition and monitoring control system into the newly designed DenseNet network,and extracts the representative features of faults through layer by layer training.In addition,the discrete eigenvalues are extended to Euclidean space by using the unique hot coding,so that the diagnosis network can distinguish the fault features more accurately.Secondly,in view of the high frequency noise of wind speed,ambient temperature and other signals in the input and the deficiency of DenseNet network in the fault early warning problem with high correlation of processing time,an intelligent fault early warning method based on the combination of improved dense connection network and gated cycle unit(GRU)is designed.In this method,the improved Bhattacharyya Distance based variational mode decomposition(VMD)is used to denoise the wind turbine operation data.After processing,the disturbance in the signal is less and the fault feature information is clearer.At the same time,the introduction of GRU also makes up for the deficiency of DenseNet,which can get more accurate fault warning effect.The improved VMD-DenseNet-GRU network realizes the prediction of fault early warning index with the help of customized dense connection module and its memory gating structure.By using sliding window and three fifths principle of fault interval discrimination,the false alarm caused by short-term meteorological changes can be greatly reduced.Finally,the above two methods are applied to the actual fault data of wind turbine to verify the effectiveness of the method.