Abstract:
For a steam power plant, condenser vacuum is an
important variable to monitor. Insufficient condenser vacuum can
cause high heat rate. An approach is proposed to modeling
condenser vacuum based on the autoregressive-moving-average
(ARMA) combined with the generalized-autoregressiveconditional-heteroscedasticity (GARCH) technique. It makes use
of data available from a generating unit of the Asam-asam Steam
Power Plant over a period when the unit was running under severe
off-design conditions. In that period, the unit experienced poor
condenser vacuum. The data contain observations on variables,
some of which are important for studying conditions regarding
condenser vacuum at the unit. The resulting models can explain
how condenser vacuum varies in response to changes in
conditions. The predictive performance is comparable to that
obtained using autoregressive neural network and support vector
regression. Further remarks on the modeling issues are given.
