Online Soft Sensor in the Main Fractionator of FCC Unit

摘要

In most industrial processes, it is difficult to measure product quality directly. To overcome this difficulty, the concept of soft sensor has been introduced and has gained widespread attention. In this paper, a soft sensor model, based on the Radial Basis Function Neural Netwrok (RBFNN), is established and the selection of RBFNN input and RBFNN training method are discussed. The soft sensor model uses online correction to track and estimate any product quality variation to the required accuracy. It was used to estimate the end-point of naphtha and freezing-point of light cycle oil for the main fractionator of the fluid catalytic cracking unit in a refinery. Real-time application results showed that its measured errors were within +-2 deg C which satisfied measured requirements. The successful application of this soft sesnor model not only saved investment costs but it also allowed for multivariable control.