Rob J Hyndman

Shu Fan and Rob J Hyndman

Revised 10 Janu­ary 2011

IEEE Trans­ac­tions on Power Sys­tems (2012), 27(1), 134–141.

Abstract
Short-​​term load fore­cast­ing is an essen­tial instru­ment in power sys­tem plan­ning, oper­a­tion and con­trol. Many oper­at­ing decisions are based on load fore­casts, such as dis­patch schedul­ing of gen­er­at­ing capa­city, reli­ab­il­ity analysis, and main­ten­ance plan­ning for the gen­er­at­ors. Over­es­tim­a­tion of elec­tri­city demand will cause a con­ser­vat­ive operation, which leads to the start-​​up of too many units or excess­ive energy pur­chase, thereby sup­ply­ing an unne­ces­sary level of reserve. On the con­trary, under­es­tim­a­tion may res­ult in a risky oper­a­tion, with insuf­fi­cient pre­par­a­tion of spin­ning reserve, caus­ing the sys­tem to oper­ate in a vul­ner­able region to the disturbance.

In this paper, semi-​​parametric addit­ive mod­els are pro­posed to estim­ate the rela­tion­ships between demand and the driver vari­ables. Spe­cific­ally, the inputs for these mod­els are cal­en­dar vari­ables, lagged actual demand obser­va­tions and his­tor­ical and fore­cast tem­per­at­ure traces for one or more sites in the tar­get power sys­tem. In addi­tion to point fore­casts, pre­dic­tion inter­vals are also estim­ated using a mod­i­fied boot­strap method suit­able for the com­plex sea­son­al­ity seen in elec­tri­city demand data. The pro­posed meth­od­o­logy has been used to fore­cast the half-​​hourly elec­tri­city demand for up to seven days ahead for power sys­tems in the Aus­tralian National Elec­tri­city Mar­ket. The per­form­ance of the meth­od­o­logy is val­id­ated via out-​​of-​​sample exper­i­ments with real data from the power sys­tem, as well as through on-​​site imple­ment­a­tion by the sys­tem operator.

Down­load work­ing paper

Online pub­lished version