Short-term And Ultra-short-term Forecasting Of Wind Power Based On Neural Network

Wind energy has shortcomings such as intermittent and large fluctuation,which will increase the output power fluctuation of wind power generation equipment and increase the difficulty of control.Eventually,it will be difficult for the actual power generation to accurately meet the power generation plan and power requirements.Constantly optimizing wind power forecasting methods and constructing a more accurate wind power forecasting model is one of the important means to deal with this problem.In this thesis,the following work is mainly done:To reduce the impact of outlier on wind power forecasting,an outlier identification and correction model based on Density-Based Spatial Clustering of Applications with Noise(DBSCAN)algorithm and linear regression method is proposed in this thesis.The model identifies the outliers in the wind speed and wind power data according to the density connection between the data points,and uses the normal data near the outliers as the sample of the linear regression method to correct the outliers.The analysis of the calculation example shows that the model can identify and correct the outliers effectively.Wind speed,wind direction,temperature,humidity,and air pressure all have an impact on wind power,and wind speed has the greatest impact on wind power In this thesis,the wind speed data are used to calculate the average wind speed,maximum wind speed difference and average wind speed change rate of each historical day.Based on these standards,DBSCAN algorithm and Euclidean distance are used to filter out similar day data as the input of wind power forecasting model.A short-term wind power forecasting model is built based on BP neural network optimized by genetic algorithm(GA-BP),which uses historical wind power and wind speed data with and without outlier identification and correction respectively,together with Numerical Weather forecasting(NWP)data as the input of the forecasting model.The analysis of the calculation example shows that the Normalized Root Mean Squared Error(NRMSE)of the forecasting with outlier identification and correction is reduced by 14.7%,and the Normalized Mean Absolute Error(NMAE)is reduced by 10.7%.The result proves that the identification and correction of outlier can effectively improve the accuracy of wind power forecastingIn order to speed up the forecasting,gray correlation analysis is used to screen meteorological factors,and finally wind speed,wind direction cosine,and temperature data that have a greater impact on wind power are selected as the input of the forecasting model.Using the historical operation data of wind power plant and NWP data,based on Recurrent Neural Network(RNN),combined with the window processing method,the ultra-short-term wind power forecasting is carried out.Firstly,one previous historical predicted power data is selected as the input of the current forecasting time by analyzing the forecasting duration and accuracy.The forecasting results are compared with those of RNN without window processing,RNN without meteorological factor screening,Back Propagation(BP)neural network and RNN without outlier identification and correction,and the analysis of the calculation examples shows that the NRMSE of the forecasting model is reduced by 58.4%and the NMAE is decreased by 71.5%compared to the forecasting without window processing.Compared with the forecasting without the meteorological factor screening,the forecasting time of the model is significantly shortened and the forecasting accuracy's difference is tiny.Compared with the forecasting of traditional BP neural network,the model's NRMSE is reduced by 45.8%and NMAE is reduced by 60.7%.Compared with the forecasting without outlier identification and correction,the model's NRMSE is reduced by 18.8%and NMAE is reduced by 28.4%.