Load forecasting based on partial mutual information and multiple kernel least squares vector regression

Load forecasting is important to power system. A load forecasting model is proposed here, which based on the partial mutual information estimation and the multiple kernel learning. The inputs of the model are selected according to the partial mutual information which can use to measure statistical dependence. The training samples are mapped into a high dimensional feature space by a nonlinear function with cooperative structure, and then fitted by a multiple kernel least square support vector regression model. The kernel matrix and the regularization parameters of this model are optimized simultaneously in a quadratically constrained linear Program. The application shows that the proposed model has higher prediction accuracy and better generalization performance than LS-SVR.