3 - Heuristic prediction schemes

Abstract

This chapter covers heuristic prediction schemes, which try to make statistical regressions “smarter” by factoring in at least some of the most important relations that determine a target output variable. These relations are almost always at the phenomenological level, rather than the mechanistic. The bulk of this chapter is based on a software product called the NOXLOI Predictor because its performance in emissions predictions has been fully validated and very well documented. The NOXLOI Predictor tightly focuses the calculation sequence on the incremental changes to NOX and LOI emissions due to a fuel switch, when all other operating conditions remain the same. The analysis is explicitly partitioned into factors that directly affect a fuel’s distinctive NOX and LOI formation tendencies, and factors that pertain to furnace operating characteristics. All factors related to furnace operation are represented by the internal database on emissions for standardized conditions and the baseline emissions input data. The fuel quality impacts are predicted from comprehensive reaction mechanisms based only on standard fuel properties via a “virtual fuels laboratory.” This hybrid approach works very well for NOX predictions but is less accurate for LOI emissions. Hybrid schemes like the ones for NOX and LOI are certainly not restricted to only these two emissions. Other commercial packages demonstrate that virtually all the fuel quality impacts in pc furnace operations can be accurately managed by some combination of statistical methods and heuristic treatments.