Research On Condition Monitoring And Fault Early Warning Of Wind Turbine Based On Deep Learning

In recent years,with the rapid development of wind power generation,the number of units and installed capacity have increased steadily,and it has become the third largest power source in China.However,with the increase of operating life,the problems such as frequent faults and poor reliability of wind turbine become more and more prominent.Maintenance after the failure of the unit will increase the difficulty and cost of maintenance,resulting in greater economic losses.The condition monitoring and fault early warning of wind turbine based on SCADA data and reminding technicians to deal with them in time will help to reduce the operation and maintenance cost of wind farm,improve the availability of wind turbine and prolong the life of wind turbine.Because of the large amount of data in the SCADA system,the deep learning algorithm can be used to mine the hidden information in a large number of data more fully,establish a fan performance prediction and condition early warning model with higher precision,wider adaptability and stronger generalization ability,and improve the accuracy of fault early warning.Therefore,the research on wind turbine fault early warning and condition monitoring based on SCADA data and deep learning has important academic and engineering practical significance.The main research work of this paper includes:(1)In order to solve the problem that the data recorded by SCADA system contain a lot of noise,a data preprocessing method based on improved Bin algorithm and dispersion analysis is proposed to prepare the data for the establishment of unit health model.(2)In order to reduce the complexity of the unit health model,the improved random forest algorithm is used to sort the characteristic importance of the unit parameters,and the reasonable characteristic parameters are selected as the input variables of the fan performance prediction model.(3)On the basis of data preprocessing and characteristic variable selection,the normal performance prediction model of the unit based on long-term and short-term memory network is established,and the training,testing and performance evaluation of the model are completed.(4)By using the exponentially weighted moving average method,the residual sequence between the output value of the prediction model and the actual value is smoothed,the evaluation index of unit performance state is constructed and the early warning threshold is determined reasonably,and finally the early warning of fault is realized.The effectiveness of the early warning method is verified by real historical fault cases.(5)With the help of the GUI design tool of MATLAB,the fault early warning and monitoring platform is developed to realize the visualization of unit condition monitoring and early warning.The research work of this paper shows that using the data preprocessing and feature screening method given in this paper to establish the fan performance prediction model using long-term and short-term memory network can effectively improve the accuracy and adaptability of the model.Based on this model,the abnormal state and fault early warning scheme designed can monitor the state of the unit in real time,find the abnormal unit in advance and carry on the early warning,which is helpful to improve the operation reliability of the fan unit.