Hierarchical Bayesian modelling of the electricity load
Résumé
In this paper, we study a non-linear model used to estimate and forecast the electricity load, that usually requires four or more years worth of data to avoid any overfitting phenomenon. We first propose a non-informative prior to be used when the number of observations is large enough. When the observations are too few, we propose a hierarchical prior to include information coming from another bigger, similar, sample. The posterior densities associated with these two priors are derived and a MCMC algorithm is provided in each case. We finally run these algorithms on simulated and real datasets ; the hierarchical prior greatly improves the quality of the model predictions.
Origine : Fichiers produits par l'(les) auteur(s)