| In recent years,air pollution,global warming and energy crisis have become important factors hindering international and domestic development.Wind power generation provides a strong power base for the rapid development of our country.Due to its shortcomings of intermittency and randomness,the change of wind resources may make the fan unable to generate electricity continuously,so it is more difficult for the wind field to be connected to the grid,and the safety and stability of power generation cannot be guaranteed.Most fans are located in harsh environments.The sudden change of wind resources and other external forces have a serious impact on the health of fans and the tolerance of related components.Therefore,it is necessary to forecast the wind power of wind turbine,evaluate the performance of wind turbine and fault warning.Monitoring the running state and fault of wind turbine can ensure the smooth operation of wind turbine power generation.In this context,data processing,wind power prediction,unit performance evaluation,yaw fault diagnosis and monitoring of wind turbines are studied by using the operation data of several fans in SCADA system.First of all,this paper conducted basic preprocessing according to the historical data of wind turbine operation,explored the operation mechanism and data form of the wind turbine,and eliminated the outliers in the wind power data.It was found that the LOF algorithm had a better cleaning effect on the outliers in the wind power data,and the data with low power generation performance of the wind turbine could be eliminated by the optimal intra-group variance algorithm.Then FCM fuzzy clustering is used to divide the operating conditions of wind turbines,and different operating conditions are analyzed from the data level.Secondly,based on the wind turbine operation data after cleaning,this paper selects variables using principal component dimension reduction and maximum information coefficient to determine the incoming variables of wind power prediction.On the basis of variable selection,four different forecasting methods are used to predict the wind power in short term.The results show that the prediction accuracy of short-term wind power using long and short term memory network(LSTM)is high after principal component dimension reduction of independent variables.After the original variables were selected based on MIC method,the prediction accuracy of short-term wind power was high in deep neural network(DNN).After 10 min resamsampling of the data,LSTM network was used to predict the wind power with asynchronous length in the future,and the prediction accuracy decreased with the increase of the prediction step size.The goodness of fit of DNN network wind power prediction was 0.9996,and the prediction accuracy was the highest.Based on the standard power data and actual power data of wind turbines,the power generation performance of 24 wind turbines at different wind speed segments was evaluated and analyzed.In addition,the identification of wind turbine yaw fault based on SCADA data is also a major task of this paper.The yaw process of wind turbine is more frequent,which tests the joint maneuverability between yaw system and related components.The SCADA yaw operation data of eight units in three wind fields were processed and analyzed.Based on the fan operation data,the faults generated in the yaw process were mined,and the fault discrimination threshold was analyzed by combining the fault situation and the fan yaw mechanism,so as to identify the unit yaw fault.And the visualization results are integrated.In this paper,the yaw fault results are analyzed,and the unit data with yaw gearbox fault information is used for verification.Finally,the causes of yaw gearbox fault of the faulty unit are analyzed by combining the distribution and timing of fault data.The research of this paper is based on the operation data of multiple wind farms and multiple units,and the experimental results can be mutually verified. |
PDF Full Text Request