Data-driven Research On Condition Monitoring And Fault Diagnosis Of Offshore Wind Turbine

At present,the wind power industry is developing rapidly.Compared with onshore wind power,offshore wind power has better wind energy resources and closer distance to the load center,has become an important participant in the energy market.However,compared with onshore wind turbines(WTs),offshore WTs have the characteristics of high failure rate,poor accessibility and long maintenance cycle.Therefore,timely condition monitoring(CM)and accurate fault diagnosis(FD)of offshore WTs are of great practical significance for improving the output of WTs and reducing operation and maintenance(O&M)costs.Based on the supervisory control and data acquisition(SCADA)data of offshore WTs,this paper employs data-driven methods to perform CM and FD on offshore WTs.The main research are as follows:(1)At present,the data-driven research of CM has subjective limitations in the selection of input features of normal behavior model.The attention mechanism is introduced in this paper to derive the association between the target modeling feature and the input features automatically,which can ensure the integrity of input features association information,avoid the subjective threshold limitation in the existing selection process of input features,and enhance the accuracy of normal behavior modeling(NBM).(2)At present,the NBM methods applied by data-driven CM research did not consider the temporal correlation in the SCADA data.In this paper,the temporal correlation of SCADA multi-dimensional time series data is effectively mined through the Gated Recurrent Unit(GRU)network,so as to reveal the trend of early failure earlier.(3)Currently,data-driven FD research is inadequate in the mining of fault sample set.In this paper,the mining of the early fault sample is realized based on CM technologies,which enables FD methods to effectively diagnose the early fault of offshore WTs.(4)Currently,despite the high accuracy of the algorithms used in data-driven FD analysis,the diagnostic results remain unreliable due to a lack of interpretability.Based on convolution neural network(CNN)and attention mechanism,an interpretable Convolutional Temporal-Spatial Attention Network(CTSAN)is proposed in this paper.By extracting the temporal-spatial features in SCADA data,the FD model can produce accurate diagnosis results and display the extracted temporal-spatial features in a human-understandable form,so as to improve the reliability of the FD results.Finally,based on the real data of Donghai Bridge offshore wind farm,the effectiveness and superiority of the proposed methods are verified.The results show that the proposed CM method can effectively monitor the status of the offshore WT gearbox under fault and normal operating conditions,and has higher accuracy and interpretability than the existing onshore WT CM methods,as well as the ability to reveal the fault trend sooner.The FD method suggested in this paper can effectively diagnose various faults of offshore WTs,which not only achieves the performance of the best FD model,but also shows the advantage of interpretability.