Ultra-short-term Wind Power Forecasting Based On Kernel Learning Method

Wind power generation is one of the main development directions of clean energy in the future.With the growing maturity of key wind power generation technologies and equipment,the scale of wind power plant continues to expand,and the proportion of wind power connected to the grid in China is also increasing.The strong non-linear characteristics of wind power generation have more and more prominent impact on the safe and stable operation of the national grid.Timely prediction of the output power of wind power generation is to alleviate electricity one of the effective methods to improve the stable and safe operation of the State Grid is to regulate the peak load and frequency pressure of the network.Therefore,it is necessary to study the method that can accurately predict the wind power and its changing trend.In this paper,kernel principal component analysis(KPCA)is used to extract the effective dimension of historical wind power data,simplify the input dimension of Kernel Extreme Learning Machine(KELM)algorithm,and then different kernel functions are introduced into a new type of feed-forward neural network to build a short-term wind power prediction model based on the actual wind power data.In order to improve and simplify the impact of subjective factors on the model,the optimization algorithm is introduced into the prediction model to make the kernel function parametric,and the quantity is improved,so that the model has better performance in prediction accuracy and generalization.The main contents of this paper are as follows:(1)Research on nuclear learning theory.First of all,based on the study of feed forward neural network,the basic algorithm of elm is studied in depth,and the idea of kernel function dimension up learning is introduced into elm algorithm to build kernel extreme learning machine with strong nonlinear signal processing ability,The influence of Gaussian kernel,wavelet kernel,polynomial kernel and linear kernel on Elm algorithm is analyzed,and the regularized least square algorithm is used to further improve the robustness of KELM.Secondly,on the basis of studying the basic principle and algorithm of principal component analysis(PCA),this paper focuses on the combination of Gaussian kernel function,wavelet kernel function,polynomial kernel function and linear kernel function with PCA algorithm.The kernel function is introduced into PCA to form KPCA,and the effect of kernel function on the extracted effective features and the prediction accuracy of the model is studied.(2)Combining KPCA with KELM,a super short term wind power forecasting model based on KPCA-KELM is proposed.KPCA-KELM wind power prediction model is constructed by reconstituting the nonlinear principal components extracted by KPCA into time series vector and using them as effective input of KELM method.In order to verify the rationality of the proposed method,firstly,KPCA-KELM is applied to the prediction of conventional chaotic time series,and the factors affecting the prediction accuracy of KPCA-KELM wind power prediction model are explored.Secondly,based on the measured data of Alberta wind farm in Canada and NREL laboratory in the United States,the prediction accuracy and generalization ability of the proposed KPCA-KELM wind power prediction model are verified.Through the comparative test,the influence of nuclear learning theory on the prediction accuracy of ultra-short term wind power prediction model based on KPCA-KELM is explored.(3)Based on the study of the influence on the prediction accuracy of KPCA-KELM wind power prediction model,in order to simplify the influence of subjective factors on the prediction accuracy of the model,the idea of optimization is introduced into KPCA-KELM wind power prediction model,that is,genetic algorithm(GA),simulated annealing(SA)and differential evolution(DE)are used to initialize the parameters of wind power model: input dimension Degree,kernel variable and regularization coefficient are optimized to form three different algorithms of KPCA-O-KELM: KPCA-GA-KELM,KPCA-DE-KELM and KPCA-SA-KELM.Firstly,KPCA-O-KELM is applied to the prediction of standard chaotic time series to explore the rationality of building KPCA-O-KELM wind power prediction model.Secondly,in order to test the ability of KPCA-O-KELM method to predict the ultra-short term wind power,KPCA-O-KELM wind power prediction model is applied to the actual wind power measured data provided by NREL laboratory in the United States.