Short-term Solar Energy Prediction Approach Based On Support Vector Machine And Trust-tech

With the penetration of renewable energy generation such as solar and wind continues to increase in power systems,the associated uncertainty and the intermittent nature of the renewable energy bring a great challenge to power system stability and security.The need for accurate solar energy prediction attracts many research and development efforts.In this study,an ensemble of TRansformation Under STability-reTraining Equilibrium CHaracterization(Trust-Tech)-enhanced,group-based genetic algorithm(GA)-assisted SVM predictors is proposed.Several distinguished features of the proposed method are as follows: Firstly,feature selection algorithm is used to choose relevant features driven by data for high prediction accuracy.This algorithm can also speed up training speed by reducing irrelevant and redundant input feature dimension.Then,a Trust-Tech-enhanced,group-based GA is proposed for deriving high-performance and diverse SVMs.This method can get full use of valuable solutions during GA evolutionary process and obtain more accurate individuals.Finally,an ensemble method of diverse high-performance SVMs is developed to further improve forecasting accuracy.In another part of this thesis,multiple kernel learning based support vector machine is proposed to predict the solar energy.Multiple kernel learning can combine statistical learning theory with domain knowledge properly and efficiently by designing different combination of multiple kernels,both linear and nonlinear,considering distinct input data to improve prediction accuracy.A part of published solar energy prediction contest(American Meteorological Society(AMS)2013-2014 Solar Energy Prediction Contest)dataset is used to evaluate the performances of the proposed method.Numerical studies along with comparison results demonstrate that the proposed method can achieve promising forecasting accuracy.