Research On Dynamic Modeling Of Mechanical Operation Of Wind Turbine

For modern megawatt-level wind turbines with multi-degree-of-freedom control capabilities,how to fully utilize the potential controllability and improve the control level becomes more and more urgent to reduce the cost per kilowatt hour of wind power and to enhance the healthy development of wind industry.Many theoretical control algorithms are faced with severe challenges when they are applied in the actual field.It is very difficult to establish a control-oriented model of the complex nonlinear dynamic characteristics of the actual wind turbine.Although some scholars have extensively studied the modeling of wind turbines,they have not yet provided effective solutions for practical applications.Therefore,this paper establishes a mixed semi-parametric model of the output from the mechanical side of the wind turbine on the basis of data preprocessing,which mainly includes the following three parts:The Supervisory Control and Data Acquisition(SCADA)system of the wind turbine stores a large amount of operating data and a lot of abnormal data is recorded,which has an impact on wind turbine related research.These abnormal data need to be identified and eliminated.First,according to the operation mechanism of the wind turbine,the abnormal data is identified,and the remaining data is clustered by Gaussian mixture model in the wind speed-rotor speed-pitch angle three-dimensional space to identify the category of the abnormal data,and then the abnormal data is performed in the Copula space Refined recognition.According to the operating conditions of wind turbines,the data under the power limit and maximum power tracking operation state are retained.Due to some reasons such as the elimination of abnormal data or operating status,there will be missing values in the data in the SCADA system,resulting in insufficient data quality and quantity to support related research.Therefore,these missing data need to be filled.Using continuous wind speed data and remaining data fragments,on the basis of similar conditions,the Gaussian mixture model and LSTM network are used to train the missing data filling model.In order to reduce the uncertainty caused by the single method and the one-way filling method,different filling methods can be weighted or forward and backward filling can be weighted to obtain the final filling result.The model based on the piecewise affine(PWA)structure can better approximate the complex nonlinear dynamic characteristics.On this basis,this paper proposes a PWA-based hybrid semi-parametric modeling method,derives its principle,for the actual operating characteristics of the mechanical side of the wind turbine,the piecewise affine include output autoregressive(PWARX)model of the wind turbine,the PWA-based mechanism model parameter identification state-space model,and the PWA-based LSTM model was separately established.For the generated errors,the LSTM is used to train the error compensation model,which greatly improves the accuracy of the model.The effectiveness of the proposed method is verified by the data generated by the actual operation of wind turbines,which provides an important reference for studying the data processing of actual wind turbines and analyzes the dynamic characteristics of operation,and also lays an important foundation for the control of actual wind turbines.