Abstract
Accurate short term electricity load forecasting is crucial for ef-
ficient operations of the power sector. Predicting loads at a fine
granularity (e.g. households) is made challenging due to a large
number of (known or unknown) factors affecting power consump-
tion. At larger scales (e.g. clusters of consumers), since the inherent
stochasticity and fluctuations are averaged out, the problem be-
comes substantially easier. In this work we propose a method for
short term (e.g. hourly) load forecasting at fine scale (households).
Our method use hourly consumption data for a certain period (e.g.
previous year) and predict hourly loads for the next period (e.g.
next 6 months). We do not use any non-calendar information, hence
our technique is applicable to any locality and dataset. We evaluate
effectiveness of our technique on three benchmark datasets from
Sweden, Australia, and Ireland.