Pattern Recognition in Industry

A refinery with a reactor unit much larger than standard units of its type was required to meet stack emission specifications. What makes this case especially interesting is that the allowed limits for the particulate emissions were on an absolute mass rate basis rather than a percentage of the stack gases; and the specification was based on that for a unit of average size, thereby making it a very challenging target for a very large reactor to meet. In order to optimize operational profits, it was essential to be able to use low cost feeds with high sulfur content without exceeding the stack emissions regulatory limits.
The goal was to model the unit's behavior in order to be able to predict the stack emissions given the feed to the unit and the operating conditions, and then to identify operating windows which would permit the profitable use of high sulfur content feeds without exceeding the limits for stack emissions.
A simplified overall flow diagram for the reactor unit is shown in Figure 13.1. The feed and a regeneration stream flow into the primary reactor. The effluents from the reactor go through off-gas boilers and scrubbers before entering the exhaust stacks. Domain experts identified potentially important input parameters that governed the unit's behavior. This list of parameters was screened for mathematical resonance with the intended outcome (the rate of particulates emitted in the stack) and was subsequently pruned to generate the following set of...