Multi-dimensional Anomaly Identification Of Wind Turbine Power Curve Based On Hybrid Copula Function

Under the background of vigorous development of new energy industry,wind power industry has become an irreplaceable and important part of new energy power generation industry.With the upgrading of the first generation wind turbines and the gradual increase of the scale of wind power in the world,higher requirements are put forward for the ability of abnormal data identification and the means of unit health assessment of high-power wind turbines.Power curve is one of the important indexes to evaluate the operation performance of wind turbines.The operating conditions of wind turbines can be more accurately distinguished by the trend of data reflected in the curve.In this thesis,based on the mixed Copula function,the abnormal data in the power curve is identified,and the health status of wind turbines is evaluated.The main work of this thesis can be divided into the following contents:First of all,in view of the difficulty in identifying abnormal data of power curve of wind turbine at present,this paper proposes a method based on hybrid Copula function to identify multidimensional anomalies of power curve of wind turbine.According to the maximum wind energy tracking phase,generator torque control phase,and constant-speed pitch control phase during normal operation of wind turbines,the power curve is divided into three working conditions.Based on this,the probability power curve model is constructed respectively,and the data points outside the upper and lower boundaries are judged as abnormal points.On this basis,the curves of wind speed-impeller speed,wind speed-generator speed and wind speed-pitch angle are established.The multi-dimensional anomaly identification of wind turbine power curve is completed.The accuracy of the identification method is verified by residual timing diagram.Secondly,because the measured wind speed is less than the incoming wind speed,the measured power curve deviates from the standard power curve.In this thesis,the measured wind speed is corrected based on the wind energy utilization coefficient,and the power curve model is established according to the corrected wind speed.According to the wind speed before and after modification,the annual power generation is estimated,and then the operation characteristics of wind turbines are characterized more accurately.Finally,aiming at the problsem of missing data in the range from the rated wind speed to the cut-out wind speed,the power curve is extrapolated based on particle swarm optimization algorithm to obtain a complete power curve.Based on this,the deviation degree between the measured power curve and the standard power curve is studied.Since the hybrid Copula function can only perform qualitative analysis on the health status of wind turbines,a method based on LSTM neural network for quantitative analysis of its deviation is proposed.By solving the deviation degree of power in each wind speed interval and completing the calculation of the weight coefficient in the wind speed interval,the deviation degree of the whole power curve can be obtained.Based on this,the health condition of wind turbine is evaluated.