Ultra-short-term Prediction Of Output Power Of Large-scale Wind Farm Based On Decision Tree Theory

With global fossil energy exhaustion,tight supply and severe climate change,developing clean energy is becoming one of the important strategies of all countries.It is estimated that by 2020,the installed capacity of wind power worldwide will reach 1.2 billion kilowatts,which can meet the demand of 12% of the world's electricity.Large-scale wind power grid connection will cause impact on the grid.Wind power forecasting is an important means to relieve grid frequency pressure and ensure stable grid operation.According to the “Provisional Measures for Management of Wind Power Plant Forecasting” issued by the Energy Administration in 2011,real-time forecasting refers to the forecast of 15 minutes to 4 hours from the time of reporting,with a time resolution of 15 minutes.Therefore,the real-time forecast of wind power in this project is based on the wind power time series with a time interval of 15 minutes as the main research object,and it will perform rolling forecasting of 16-step ultra-short-term wind power forecast.The forecast results obtained from this method can serve to adjust the real-time active output of the wind farm units,which is of great significance for improving the utilization of wind energy.This article focuses on the characteristics of wind power data.Perform wind-power curve feature analysis.Analyze the influence of wind speed fluctuation characteristics and wind direction factors on the power output of wind turbines.The data of different wind characteristics were respectively fitted with the wind speed-power curve to obtain the characteristic wind speed-power curve.Wind power actual measured historical data is an important basis for wind power modeling and predictive analysis.However,there are abnormal data points in the measured historical data of wind power.These data do not reflect the actual operating characteristics of the wind turbine,affecting the effectiveness of model establishment and affecting the forecast results.This article analyzes the timing characteristics of abnormal data.Use the Copula function to establish a probability power curve.Combined with the characteristic wind speed power curve described in the previous chapter,an abnormal data identification algorithm was established.Based on the support vector machine algorithm,a data reconstruction model was established based on the characteristics of wind speed and elevation characteristics and the strong correlation between neighboring units.Guarantee the validity of the data.This article analyzes the advantages and disadvantages of a single predictive model,and analyzes the characteristics of the predictive results of two models based on historical data and models based on NWP data.Propose a combined predictive model that combines the advantages of two single models.Analyze the characteristics of each sequence in the historical data set and the corresponding best prediction method.The decision tree is used to build the classification model,and the optimal forecasting method is matched according to the characteristics of real-time data,and power forecasting is performed.An example is analyzed by the historical data of a wind farm to verify the validity of this model.