What is Data Mining 2012
Suppose you wanted to optimize a cyclone furnace (an older-type design for burning coal, still in
use in many power plants) for stable high flame temperatures. Stable temperatures are necessary
to ensure cleaner combustion, and less build-up of undesirable slag that may interfere with heat
transfer. Typically, most power plants are equipped with very effective data gathering and
storage technologies, so there are easy ways to extract the data that describe various parameter
settings, as well as flame temperatures, on a minute-by-minute interval.
Traditional methods to approach this task – to optimize combustion to achieve stable flame
temperatures in the presence of different loads, fuel quality, and so on – come down to the
application of a-priori (CFD) models, or more or less trial-and-error parametric testing.
CFD (Computational Fluid Dynamics) modeling
One approach is to use explicit theoretical (first principles) models, to understand (based on
these usually complex and highly nonlinear models) how best to set certain parameters, distribute
airflows, etc. to optimize performance. With an explicit theoretical knowledge (model) of how
exactly various parameters affect flame temperatures, one can use standard computer
optimization algorithms to identify optima, which "in the laboratory" can be expected to
optimize for stable flame temperatures.
Typically, these methods are used to identify the parameter "boundaries" where to keep certain
input parameters (controlled by operators, or closed loop control systems) to ensure stable
operations. However, in practice, there are numerous obstacles that put limitations on the
applicability, effectiveness, and usefulness of CFD methods to optimize furnace performance "in
vivo", i.e., inside a "real" power plant.
First, theoretical, a-priori, physical models of furnaces will only model parameters that are
known (consistent with models) to have an influence. If in a particular installation, there are
other specific "noise factors" that effect performance, CFD will not "know about this", nor can
CFD models accommodate various esoteric installation details in a real power plant.
StatSoft White Paper July 2012
Second, CFD models can be very complex, and indeed become practically impossible to
optimize because of their complexity.
So what is often needed is a "simplification" of sorts, or a "proxy-model" ("stand-in") that can
summarize how the parameter inputs such as over fired air (OFA) distribution, primary and
secondary air flows, coal-flow, and so on will affect flame temperatures, and the variability in
flame temperatures. Data mining methods can provide such "proxy models", as will be further
explained later.
Continue Reading
