engleski

Development of Soft Sensors for Debutanizer Product Quality Estimation and Control

Law regulations dictate firm restrictions of product quality specifications and refinery emissions. Measurement of great number of process variables and installing new expensive process analyzers is necessary for efficient process control. Possible solution of this problem is application of soft-sensors. This paper demonstrates soft-sensor design for product quality monitoring and process control of debutanizer column of INA Refinery Sisak, Croatia. The column is fed by unstabilized FCC gasoline, and products are Liquefied Petrol Gas (LPG) and stabilized FCC gasoline. Method of estimation of pentane fraction in liquefied petrol gas (LPG) and Reid vapor pressure of stabilized FCC gasoline using inferential model is elaborated. The aim is to control debutanizer thus pentane fraction in LPG is kept under 2 mass percent and RVP of FCC gasoline on desired value (50 kPa). Two neural soft sensor models are developed based on available process measurements and laboratory analysis – first for estimation of pentane fraction in LPG and second for estimation of RVP of stabilized FCC gasoline. Temperatures on the several trays and reflux flow rate serve as inferential variables. For the building of the neural networks the cascade learning based on the cascade-correlation learning paradigm is developed. Developed soft sensors have been validated by additional experimental data and achieved results have been analyzed and compared with laboratory analysis results. Neural network-based soft sensors are shown to be a good alternative to hardware analyzers for debutanizer products and can be built by using data from existing plant. Also, they make possible continuous product quality monitoring and process control.

proces modeling; soft sensor; neural network; debutanizer column

nije evidentirano