Very short-term load forecast for demand side management in absence of historical data

Abstract

Load   forecasting   plays   an   important   role   in today’s  electricity  grid,  due   to   the   presence   of   distributed generation.  In  this  paper,  various  methods  for  short-term  load forecasting  are  presented  with  the  purpose  to  serve  as  tool  to operate within a demand side management system. The need for large   amount   of   historical   data   is   avoided,   opening   the possibility  for  immediate  use  by  customers  without  recorded load  history.  Widely  distributed,  such  system  can  increase  the reliability  of  the  modern  grid.  The  results  showed  that  the  Holt-Winters  forecasting  procedure  give  better  results  in  comparison to  neural networks  with  gradient  descent,  principally  if  the  load rises fast in the early morning hours.