Application of canonical variate analysis to the dynamical modeling and control of drum level in an industrial boiler

The paper describes an application of canonical variate analysis (CVA) and 9-element LQG-optimal control (linear quadratic Gaussian) to boiler drum level control on a 400 kpph gas fired power boiler. Consumed air and flue gas temperatures are investigated for their value as clean fuel flow proxy variables. The work thus has relevance for drum level control of dirty fuel boilers. The refined model selection capabilities of CVA and the Akaike information criteria (AIC) are demonstrated on this challenging process. While LQG-optimal control appeared to be more intelligent than 3-element control, the desired objective of reducing drum level excursions to 50% of "best" 3-element control was not achieved for fireside disturbances. Suggestions for improving model quality are included.