A Comparison of Multiple Methods for Short-Term Load Forecasting

We present five methods for probabilistic short-term load forecasting, namely, Bayesian estimation, a rank-reduction method based on principal component analysis, least absolute shrinkage and selection operator (Lasso) estimation, ridge regression, and a supervised learning approach called scaled conjugate gradient neural network. We aim to incorporate the load and temperature effects directly in these methods to reflect hourly patterns of electrical demand. We provide empirical results based on data sets used in the Global Energy Forecasting Competition 2014 and show that ridge regression has a marginal advantage over the other methods.