摘要：AbstractThe bottoms product of a debutaniser column in a Fischer-Tropsch refining catpoly unit should be maximised to ensure optimal operation of the downstream units. An accurate estimate of the C4 hydrocarbons in the bottoms product is required to ensure that the specification is not violated. This work demonstrates a practical implementation of a soft sensor to estimate the %C4 material in the bottoms product of the debutaniser using the General Distillation Shortcut (GDS) method and a random forest (RF) machine learned model. The paper highlights practical challenges when deploying a soft sensor to an industrial plant. It is shown how the GDS method soft sensor had to be refitted after unit maintenance was carried out. In comparison the RF model soft sensor uses more reliable measurements and did not require refitting after unit commissioning. Both soft sensors performed well and the choice of soft sensor depends on the available measurements and measurement reliability.
关键词：Keywordsindustrial applications of process controlmachine learningpetrochemical refiningrandom forestssoft sensor
