| With the increasing wind power penetration in power systems,the inherent uncertainty of wind power brings severe challenges to the safe and economical operation of the grid.Statistical analysis of wind power output probability characteristics and implement of very short-term wind power prediction with high accuracy can provide the essential fundermentals for electrical power system operators and managers to deal with wind power uncertainty.The higher the prediction accuracy is,the stronger the ability of the grid to accept wind power will be.On the other hand,more replacement of conventional synchronous generators by wind turbine generators gradually deteriorates the frequency response characterisics of power system.Therefore,the participaton of wind turbine generators in frequency regulation of power system earns widespread recognition both in industrial and academic community.Based on the measured data of a real wind farm located in Sichuan Province,this project conducts research on wind power output characteristics,very short-term wind power prediction and inertial control,detailed contents are as follows:1.Statistical analysis of wind power distribution and wind speed distribution of the wind farm of one year are conducted,and the wind power fluctuation characteristics under 15min,30min and 60min time scales are compared.Correlation and complementarity of different wind turbine generators in the wind farm are analyzed.Meanwhile,analysis of the large range of distribution of wind power output of single wind turbine generator under the same wind speed is carried out,a strategy of wind power curve modeling based on optimal smoothing order is proposed.Utilize optimal smoothing order to preprocess the origin wind speed data to achieve the input wind speed and then the wind power curve model is established,and compare accuracy with the existing methods which utilize origin wind speed data as input.2.A completing method of wind power data loss under long time scales based on back propagation neural network(BPNN)is proposed.Utilize the measured wind power output of other seven wind turbine generators under the time periods of data loss as inputs of BPNN,then achieve the wind power data of the wind turbine generator which lost its wind power data.Accuracy comparison is carried out with the conventional adjacent wind turine method.3.The research on very short-term wind power prediction based on historical wind power data is intensively carried out.Three time series methods named persistence method,ARMA and ARIMA are implemented.Considering the fluctuation characteristics of wind power series,an enhanced persistence method is proposed,which improves the prediction accuracy to a certain extant compared with the conventional persistence method.Four intelligence algorithms named BPNN,RBFNN,SVM and PSO-SVM are implemented.Compare and analyze the prediction results of the methods mentioned above under different seasons and prediction steps.Choose the best performed time series method named ARMA and the best performed intelligence algorithm named PSO-SVM to be individual methods in the combined methods based on ANFIS,and then the final wind power prediction results are achieved.Compare and analyze the prediction accuracy of the combined method and two individual methods.4.Effects of wind speed fluctuation in transient process on system frequency regulation are analyzed,and then a very short-term wind speed prediction based inertial control strategy is proposed.Design the ROCOF loop gain and droop loop gain based on the predicted 10-second average wind speed,and update the loop gains every 10 seconds.In the simulation system,based on several wind speed fluctuation conditions from measured data,implement case research including generator tripping disturbance and load increase disturbance.Compare and analyze the control performance between the proposed very short-term wind speed prediction based inertial control strategy and the inertial control strategy with constant loop gains. |
PDF Full Text Request