Development of advanced Gaussian Process for LMP forecasting

Abstract

In this paper, an advanced Gaussian Process (GP) model is proposed for electricity price forecasting. This paper focuses on forecasting of LMP (Locational Marginal Price) that maintains the efficiency in power markets in a sense that the transmission congestion is alleviated. The power market players are interested in maximizing the profits and minimizing the risks by selling and purchasing electricity. As a result, an efficient method is required to forecast LMP and evaluate the uncertainties effectively The proposed method makes use of the hybridization of GP, DA clustering and EPSO. GP is an extension of SVM in which hierarchical Bayesian estimation is used to deal with the uncertainties of electricity price forecasting through the error bar. DA clustering of global optimization is used as the prefiltering of GP in a way that GP is constructed at clusters obtained by the clustering method. EPSO is employed to improve the accuracy in MAP estimation for GP. In addition, the Mahalanobis kernel is introduced into GP to enhance the model generalization ability. The proposed method is successfully applied to real data of hourly LMP.