Research On Forecast And Correction Of Short-term Wind Power

As environmental pollution becomes more and more serious,the consumption of fossil fuels and other fuels is increasing day by day,and the huge reserves of renewable clean energy wind power have attracted widespread attention worldwide.The use of wind power prediction technology to achieve accurate prediction of wind power power is beneficial to the online bidding of wind power,reducing wind abandonment,and alleviating the adverse impact of wind power grid integration on the grid.At present,the combined forecasting model is a hot spot in the research of wind power forecasting.However,how to choose a single forecasting model and a combination method,and at the same time,how to improve the short-term rolling forecast accuracy of the wind power forecasting model is worthy of further discussion.Focusing on the above issues,the thesis has carried out related research on short-term wind power forecasting.The main research contents are as follows:In view of the missing or abnormal problem of the original collected wind power data,the original wind power data is preprocessed.According to the rationality test of relevant wind power data,the quartile method is used to process abnormal data,to fill in missing data and modify abnormal data to make it conform to the standard wind speed-power curve of wind turbines,improve data quality,and provide modeling Accuracy provides basic guarantee.Aiming at the problem of low prediction accuracy and slow prediction time of a single model,particle swarm optimization is used to optimize the Generalized Regression Neural Network(GRNN),and the PSO-GRNN prediction method is established to predict and analyze wind power data.By analyzing the prediction effect diagrams and prediction evaluation indicators of each algorithm,comparing the PSO-GRNN model with the traditional PSO-BP model,it is found that the prediction curves are similar to the real value fluctuation trend,and the prediction effect of PSO-GRNN is better than that of PSO-BP Neural network has the best prediction performance.PSO-GRNN is selected to lay the foundation for establishing a combined prediction model.Aiming at the problem of low prediction accuracy of PSO-GRNN for high-dispersion data,a new method based on Empirical Mode Decomposition(EMD),Particle Filter(PF)and Generalized Regression Neural Network(GRNN)is proposed.First,EMD is used to decompose the wind power series to reduce the impact of data non-stationarity on the forecasting model;then,PF is used to analyze and process the decomposed high-dispersion data,and PSO-GRNN is used to analyze and process the low-dispersion data,Finally,the final predicted value of wind power is obtained by linearly superimposing the prediction results of each component.Aiming at the problem of the component with poor prediction effect after decomposition of the combined model,a method based on the time-division power prediction data correction method is proposed to correct the component with poor prediction effect.The 24 hours a day is divided into 8 segments,each segment is 3 hours,that is,6 sampling points;the cumulative number of errors is reduced to 8,compared to the previous cumulative number of errors of 48,and the revised forecast error is lower.At the same time,it is compared with the prediction effect of the original combination model.The revised combination model has higher prediction accuracy and better performance.