The machine learning of rules for combustion control in multiple burner installations

A rule-based system for the control and optimization of combustion in multiple-burner furnaces and boiler plants is described. The rules were elicited from energy engineers, coded into Prolog, and tested on the furnace of a continuous annealing line for rolled steel. The performance of one of the rules in particular is focused on and the possibility of its automatic improvement or replacement is addressed. A generic algorithm for learning control rules is introduced and its use in creating rules on simulations of multiple-burner installations is described. The performance of the automatically generated rules is then compared with that of the rule acquired from the experts. It is shown that the particular rule generated for each simulation shows comparable performance to that of the expert-elicited rule. However, the performance of the rule generated over all the simulations simultaneously, is not as good as the expert rule