tion process in our circulating fluidized
bed boiler (CFB). This effort began in 2003
when a graduate student, working at the
UI power plant water lab, asked if he could
use some power plant data for a summer
class project on data mining. This led to
a partnership with the UI College of Engineering Intelligent Systems Laboratory on
our data mining combustion optimization
(DACOMO) project. The project is sponsored by the Iowa Energy Center (www.
energy.iastate.edu).
The power plant’s CFB burns a combination of coal and biomass (oat hulls).
Air is added to the combustion process
using primary air in the bottom of the
combustion chamber and secondary air
above the lower bed. The ratio of primary
to secondary air affects combustion efficiency and varies with boiler load and
coal/oat hull blend ratio.
CFB boiler combustion efficiency is
calculated automatically by dividing the
boiler heat output (steam enthalpy x mass
flow) by the metered fuel (fuel weight x
heat content) input to the boiler. CFB
process operating data are available in
the data historian for a wide variety of
boiler loads, fuel mixes, air flows and air
ratios. Historian data is mined, and results
are used to construct a process model
based on past boiler performance.
Results of the data mining are
expressed as rules in the form of ‘IF, THEN’
statements. The following is one hypothetical example of a learned rule produced by
data mining: IF (oat_hull_heat= 50 and
coal_heat= 50 and boiler_master=85 and
PA_flow=103 and SA_flow=72) THEN
(boiler_efficiency=86). The data mining
process also produces information that
reveals validity and strength of the rules
it discovers. As can be seen from the
example, multiple process variables and
setpoints may exist in an individual rule.
The real power of the DACOMO system
can be seen, for example, when the system
monitors the process and automatically
recommends process setpoint changes to
improve boiler combustion efficiency. UI’s
research on this system has progressed
from lab study and simulation to actually
making recommendations to boiler operators, communicated via a Web site accessible only to the lab and operators.
As shown in figure 2, from a sample
DACOMO Web page screen, DACOMO is
recommending the operator change where
the process is operating both primary and
secondary air flows, as well as boiler draft
and the bottom ash setpoint – a total of
four process parameters. The operator
will then have to decide whether or not
to accept these recommendations and
how to change the process to obtain the
recommended operating points.
The UI’s next step in the DACOMO
project will close the automated loop and
allow the DACOMO system to directly
change process air bias settings. The human
operator will assume a supervisory role
and only interfere if the process begins
to experience upset conditions, or if it is
necessary to stop setpoint change recommendations for evolutions such as soot
