Peak load models are not widely used since they do not provide any
information about the shape of the load curve. Load shape models
describe the load as discrete time series over the forecasting interval
and can be categorized into two groups: static models 15-71 and
dynamic models. Dynamic load shape models incorporate in their
predictions the cumulative effects of such factors as recent load
behavior, weather, and random effects. They are of two basic types:
autoregressive moving average (ARMA) models [8-101 and state
space models [ 111.
Finally powerful regression models have been developed that utilize
nonlinear transformations, reverse-error-in-variables techniques
and other statistical methodologies to effectively capture load
variations due to special events, weather pattern deviations from
normal and other random correlation effects [12]. However,
inherent limitations exist in most conventional statistical methods
because of the following major drawbacks: 1) the nonlinear
relationships of the input and output variables are difficult to
capture; 2) the collinearity problem of the exploratory input
variables limits the number of input variables that can be used in
the model; and 3) the models are not very flexible to rapid system
load changes