|
Descripción
|
|
|---|---|
| This work describes a time series model used by the Spanish System Operator (Red Eléctrica de España, REE) to make hourly forecasts of electricity demand from one to ten days ahead. This article presents a new approach for estimating multiequation models for hourly load forecasting that extends previous work in diferent important ways. In our version, the 24 hourly equations are assembled to form a periodic autoregressive moving-average model. This adjustment not only improves the short term predictions for the coming hour but, its effect is noticeable beyond 24 hours, horizon of enormous practical interest in the programming of an electrical system. The model considers the hourly series as a periodically correlated process. It incorporates temperature as an explanatory variable and takes into account the effect on the demand of holidays. Direct estimation of the model by maximum likelihood is very complicated because it contains a large number of parameters. The problem is solved with a two-step estimate: ?rst, an independent model is obtained for each hour (this is the usual strategy in most of the models proposed in the literature) and, secondly, a joint estimate of the parameters that relate each hour to the previous hours. In this work it has been applied the methodology of regression splines to the problem of estimating the functional relationship between weather and electricity demand. The nonlinear relationship is observed graphically. The results agree with intuitive expectations, and the graphs clearly show how temperature in?uences demand changes throughout the day. The dynamic effect of the temperature on demand is incorporated by simply adding lagged temperature variables. The choice of the number of nodes is performed by cross-validation;in this implementation 3 nodes are used. These have been equispaced. It has been proven experimentally that the overall behavior of the method is quite robust. | |
|
Internacional
|
Si |
|
Nombre congreso
|
37th international symposium on forecasting |
|
Tipo de participación
|
960 |
|
Lugar del congreso
|
Cairns, Australia |
|
Revisores
|
Si |
|
ISBN o ISSN
|
ISSN 1997-4124 |
|
DOI
|
|
|
Fecha inicio congreso
|
25/06/2017 |
|
Fecha fin congreso
|
28/06/2017 |
|
Desde la página
|
70 |
|
Hasta la página
|
70 |
|
Título de las actas
|
ISF - Proceedings |