Short-Term Wind Power Forecasting Based On Probability Kernel Method

With the gradually mature of wind power technology and the continuously increasing of grid wind farm, the impact of wind power generation on power grid increasingly apparent, and wind power prediction is of great significance to the development of power system planning. However, with the impact of climate the nonlinear and uncertain natural of wind resources making the short-term wind power difficult to predict. Current methods such as neural networks and support vector machines have achieved a good prediction.However, at present, the majority of the studies only focused on the expectations of point prediction technology, because the uncertainty of the wind power error is inevitable and meaningful, therefore, while improve the prediction accuracy, if you can give the scope of the prediction uncertain error, it can helps assess the decision risk of predicted results. Meanwhile assuming the mean and covariance matrix of the distribution which generated forecasting model are known, the kernel minimax probability machine regression(KMPMR) method regards the classification hyperplane of kernel minimax probability machine classification(KMPMC) as the output of the regression model and maximizes the minimum probability that the output of the model will be within some boundary of the true regression function, this method not only predicted the output but also predict the range of error distribution.Moreover, feature extraction methods kernel principal component analysis(KPCA) can preprocess the data, and it can effectively extract the input of the nonlinear principal components in the feature space.For short-term wind power, a probability forecasting method based on KPCA-KMPMR is proposed.The main contents of this thesis are as follows:(1) This dissertation research on the basis principle and algorithm implementation of principal component analysis(PCA)?kernel method and KPCA.(2) This dissertation research on the minimax probability machine classification(MPMC) methods. Then extend the MPMC method to regression problem, obtain minimax probability machine regression(MPMR) method, learn the algorithm of MPMR method.Combine MPMR method and kernel method, this dissertation obtain two KMPMR methods which are based on two different algorithms, then research the two methods of learning algorithm.Combining PCA, KPCA method with KMPMR, given PCA-KMPMR, KPCA-KMPMR methods.(3) In order to verify the effectiveness of the proposed methods, apply them to short-term wind power in different regions, under the same conditions, the prediction accuracy of the proposed method is higher than single prediction method and support vector machine(SVM), in addition, different kernel functions are applied to the proposed prediction method, verified the effectiveness of the proposed methods.