Research On Wind Turbine Generator Over Temperature Fault Warning Method Based On SCADA Data

As the size and height of wind turbines increase,it makes their maintenance costs rise significantly.Over-temperature faults are a common problem during Wind turbine generators operation and are often a precursor to the occurrence of Wind turbine generator failures.If early warning of such faults can be achieved,timely maintenance strategies can be developed to reduce the O&M costs of wind turbines.By setting dynamic thresholds for generators over temperature warning through relevant technical methods,the operating status of wind turbine generators can be accurately determined.Most wind farms are equipped with SCADA systems to monitor the operation status of wind turbines,and the wind turbine fault warning method based on SCADA data has inherent advantages.Therefore,this thesis conducts research on wind turbine generators over temperature fault warning method based on SCADA data,and the main contents are as follows:(1)Wind turbine SCADA abnormal data processing.Due to the influence of wind speed and wind direction uncertainty,a large amount of abnormal data is stored in the wind turbine SCADA system.Firstly,based on the significant difference between the frequency of normal and abnormal data in the wind speed-power diagram,an isolated forest abnormal data processing method is proposed to eliminate the discrete points that are sparsely distributed and far from the wind speed-power main band data.Then the least squares method combined with the Bin method was considered to fit the wind speed-power curves,and the anomalous data closer to the wind speed-power main band data were eliminated by3-Sigma standard interval estimation according to the difference between the fitted curves and the wind speed-power data,and high-quality healthy data were obtained to provide a data basis for the subsequent study.(2)Generator temperature field simulation study of the wind turbine.Due to the lack of wind turbine start-up temperature rise data samples in SCADA data,a generator temperature field simulation method is used to predict the temperature rise change of the generator during the start-up phase.Firstly,a two-dimensional generator magnetic simulation model and an external load circuit are built to calculate the loss values of each generator component.Then the thermal basis properties,heat dissipation coefficient,and heat generation power of generator materials are calculated to build a 3D generator temperature field simulation model.Set up the meshing method and calculate the transient temperature rise data of each component of the generator.The stator end surface temperature rise data is extracted,and the sample size is expanded by fitting the data with the third spline interpolation method,and the fit is verified with the actual data to prepare for the warning threshold setting in the start-up phase.(3)Research on deep learning model of wind turbine generator temperature data.Due to the existence of a huge amount of data samples of wind turbine continuous operation phase temperature changes in SCADA data,the method of building a deep learning model is used to predict the temperature changes of generators in the continuous operation phase.The pre-processing operation of SCADA data parameter selection and normalization is performed first.Then a many-to-many model prediction strategy and two input data construction methods are proposed.Finally,CNN-GRU-AM and CNN-LSTM-AM deep learning models are built for comparison study,and the optimal combination of deep learning models and input data structures is selected for temperature prediction.Fitting validation with actual data is performed to prepare for the warning threshold setting in the continuous operation phase.(4)Early warning threshold setting.Firstly,the residuals between the predicted and real data of the two models are calculated,and for the problem of extreme values of the residuals,a method based on the exponentially weighted moving average is proposed to smooth the residuals.Then,based on the mean and standard deviation of the residuals,the warning thresholds of the wind turbine generator start-up phase and continuous operation phase are calculated.Finally,the generator over temperature fault warning is verified and the feasibility of the method is demonstrated.