Based on the on-line learning RBF network model

a combustion performance optimization model used for boilers has been established.Through initiation of the network model

on-line detection of errors

and on-line learning

the real-time matching of the model with the combustion performance in boilers has been realized.When the combustion performance in boiler is changing

the dormant nodes in hidden level of RBF network will be activated

and take new learning sample as the center of newly added working nodes

the combustion performance model will be rapidly corrected according to the newest data on-line in the combustion process of boiler.Taking subcritical 300 MW unit for heat-supply as example

the combustion performance model of boiler has been established

and the combustion efficiency of boiler being predicted

it tallies better with the result of practical operation.