Research On Short-Term Wind Power Prediction Based On Quadratic Decomposition And Ensemble Feature Selection

With the promotion of dual carbon goals and the transformation of China’s energy structure,the capacity of wind power grid connection is increasing.The randomness,volatility and discontinuity of wind power output have brought severe challenges to the safety,economy and high-quality operation of the grid.Accurate wind power prediction is of great significance to the generation dispatching of wind farms and the peak shaving and frequency modulation of power grids.Based on this,this paper studies short-term wind power prediction,and proposes a wind power prediction model based on "quadratic decomposition integrated feature selection IPSOLSTM".The main research contents are as follows:(1)Aiming at the problem of mode aliasing of decomposed signals and the randomness and complexity of residual terms,a secondary decomposition method based on VMD-EMD is designed.The empirical mode decomposition(EMD),ensemble empirical mode decomposition(EEMD)and variational mode decomposition(VMD)methods are introduced to reduce the complexity of time series data,and the secondary decomposition algorithm is used to suppress mode aliasing.The example shows that the method can not only suppress the mode aliasing problem,but also reduce the reconstruction error on the basis of reducing the data complexity.(2)An integrated feature selection method based on fuzzy type ⅰ is established.Aiming at the problem that the traditional fusion multi feature selection method will reduce the characteristics of wind power,an integrated feature selection method based on fuzzy type Ⅰ is designed.This method combines the advantages of maximum information coefficient(MIC),Pearson correlation coefficient(PCC)and distance correlation coefficient(DC),and reduces the impact of the three feature selection methods on the difference in judgment of factor redundant features through fuzzy evaluation,The problem that redundant features will reduce the prediction accuracy and calculation efficiency of the model is solved,and the influencing factors of wind power are effectively extracted.(3)Combining the decomposition algorithm,machine learning algorithm and deep learning algorithm,a wind power prediction model of secondary decomposition integrated feature selection IPSO-LSTM is established.A series of subsequences obtained after secondary decomposition are established as base learners.IPSO optimizes the parameters of each component base learner,and finally integrates the prediction results of each component.In this paper,two completely different types of wind farms,a wind farm in northern Shaanxi and an offshore wind farm,are verified by an example,and experiments are designed to verify the prediction accuracy and robustness of the prediction model by comparing single prediction models,comparing different decomposition methods,and comparing multiple prediction periods.