A statistical model for natural gas standardized load profiles

We present a statistical model for construction and application of standardized load profiles. Standardized load profile curves give a typical natural gas consumption pattern throughout a year in various time resolutions. Our semiparametric regression model uses three types of information, constant characteristics of an individual customer, individual historical consumption and time varying explanatory variables, both to describe typical trend and to correct for departure of current conditions from normal. The model's multiplicative structure allows for convenient separation of individual-specific and common time varying parts. Although corrections are parametric, no substantial information about the typical consumption trend is available, so it is modelled non-parametrically. Corrections for temperature effects are non-linear; hence we deal with them through an obvious extension of the generalized additive model framework. Copyright (c) 2009 Royal Statistical Society.