Forecasting the short-term load entails the construction of a model, a
nd, using the information available, estimating the parameters of the
model to optimize the prediction performance. It follows that the more
closely the chosen model approximates the actual physical generating
process, the higher the expected performance of the forecasting system
. In this paper it is postulated that the load can be modeled as the o
utput of some dynamic system, influenced by a number of weather, time
and other environmental variables. Recurrent neural networks, being me
mbers of a class of connectionist models exhibiting inherent dynamic b
ehavior, can thus be used to construct empirical models for this dynam
ic system. Because of the nonlinear dynamic nature of these models, th
e behavior of the load prediction system can be captured in a compact
and robust representation. This is illustrated by the performance of r
ecurrent models on the short-term forecasting of the nation-wide load
for the South African utility, ES-KOM. A comparison with feedforward n
eural networks is also given.