Wind Power Output Of Combined Models At Different Spatial Scales Short-term Forecasting Research

Under the strategic goal of achieving "carbon neutrality," renewable energies such as wind power and photovoltaics have been rapidly developing in recent years,resulting in an increase in their proportion within the power system.Wind power,as a key component of renewable energy,is subject to randomness and volatility,particularly with the increase of its connected ratio to the grid,thereby presenting significant challenges to the safe and stable operation of the power system.Accurate wind power prediction plays a vital role in power grid management,improving wind power utilization rate,enhancing power quality,and promoting the growth of wind power industry.Wind power output prediction at different spatial scales serves varying purposes and functions.This study focuses on the prediction of wind power output at different spatial scales using a combined model for short-term forecasting.The specific research objectives are outlined below.The first objective of this study is to identify and reconstruct raw wind power data via a pre-processing model consisting of the quartile method,K-means clustering algorithm,and random forest algorithm.Missing data is initially analyzed,followed by dividing the power and wind speed parameters into several intervals.Using the quartile method,transverse and longitudinal abnormal wind speed and power points are then removed to eliminate any obvious discreteness.The K-means clustering algorithm is subsequently applied to eliminate any nearby discreteness.Finally,the random forest algorithm is used to interpolate and reconstruct abnormal data,creating a continuous and complete dataset.The second objective of this study is to propose an adaptive noise-complete ensemble empirical mode decomposition(CEEMDAN)-Bayesian optimization(BO)-long and short-time series network(LSTNet)wind power prediction model to forecast single wind power unit output.CEEMDAN is initially employed to decompose processed data into several sub-sequences to mitigate the complexity of prediction.Each sub-sequence is then input into the LSTNet model,and the BO algorithm optimizes the model parameters.The prediction results of each sub-sequence are obtained and reconstructed by superposition,producing the final prediction result.Results indicate that the model is effective,as it improves the precision of power prediction for a single wind turbine unit.When the step size is 1,NRMSE,NMAE,and R2 are 7.17%,5.07%,and 96.83%,respectively.Lastly,a combined model based on convolutional neural network,long and short-term memory network(CNN-LSTM),and support vector regression(SVR)is proposed for wind power cluster prediction.Correlation analysis is initially performed to determine clustering indicators beyond power.The K-means++ clustering algorithm is then employed to cluster wind turbines(wind farms).Each clustering index is weighted according to correlation size,and the final clustering result is determined by a weighted method.The output and characteristic data of various wind turbines(wind farms)are input into the CNN-LSTM and SVR models to obtain the prediction results of the same model in different categories.The test set is divided into two equal halves,with entropy weight and the combined model used to determine final prediction results.Two cases of large and regional wind farms are analyzed,and results indicate that this method can significantly improve prediction accuracy,particularly when meteorological characteristics of cluster individuals differ greatly.