Short-term Wind Power Forecasting Based On The Maximum Correntropy Criterion And Kernel Principal Component Analysis

In the background of fossil energy deteriorating environment and increasingly dried up,the new energy sources such as wind power become the trend of the development of the energy in the future.The characteristics of strong intermittency and big volatility of the wind energy resources may have adverse effects to the real-time operation of power grid such as the power balance,scheduling optimization and the stability of the power grid,so it is necessary to grasp the power output and change trend of wind generators(field)in advance through the reasonable and effective prediction.Aiming at the high volatility and weak Gaussion distribution feature of the wind power output,this paper proposes a new criterion——Maximum Correntropy Criterion(MCC)to guide the parameter optimization process and studied the least squares support vector machine(LSSVM)short-term wind power prediction method based on the MCC.In order to reduce the bad influence of the inappropriate input variables to the prediction,use kernel principal component analysis to extract principal components of input variables before the input prediction model.On this basis,formed the short-term wind power prediction algorithm,including the normalization of the measured data,determine the best dimension of input variables,kernel principal component analysis of the input variable,parameter optimization based on MCC(separately use grid search and particle swarm optimization),LSSVM prediction module and evaluate the result with four assessment criteria.According to the engineering requirements of province electric power company,developed wind power prediction software based on C/S structure.Finally,use the wind field data for a year to test the method,the result shows that the parameter optimization process with MCC is more responsive with the wind power output character and the prediction method and software is efficient,so the prediction accuracy could be improved about 5%-20%compared with the traditional parameter optimization method.At the same time,the prediction model joined the kernel principal component analysis can make the forecast error further reduced by more than 10%compared with the prediction model without kernel principal component analysis...