Abstract
Because of the French electricity market
deregulation, Electricité de France has to experiment new load
forecasting models, more adaptive than the operational ones. A
statistical framework like Generalized Additive Models allows us
to integrate both a regressive part with explanatory variables and
an autoregressive part with lagged loads. The French electricity
demand being strongly related to the current instant, we consider
twenty-four daily time-series and fit one model for each hour. The
selected variables are one-day-lagged loads, weather and calendar
variables, and a global trend. Thanks to a cyclic spline fitted on
the position of the current day during the year, we can model the
summer break (a large decrease in the demand due to summer
holiday). We compute the Root Mean Square Errors over one
post-sample year to assess its accuracy for one-day ahead
forecast. Our model, which is fitted over five years, can compete
with the operational one.
Users
Please
log in to take part in the discussion (add own reviews or comments).