Modeling of Fluid Catalytic Cracking System Based on Improved ELM-Hammerstein Model

The complex process mechanism of fluid catalytic cracking (FCC) and the strong nonlinearity of multi-variable coupling bring significant challenges to the model predictive control (MPC) of FCC production process models. This study presents a modeling method of FCC system based on improved extreme learning machine Hammerstein (IELM-Hammerstein). Aiming at the nonlinear characteristics of FCC process, the Hammerstein model's static nonlinear subsystem is constructed using an extreme learning machine (ELM). Considering the instability of the algorithm caused by random weights and thresholds in the ELM, the sparrow search algorithm (SSA) is adopted to calculate the parameters, then one recursive least square (RLS) method is derived from calculating the linear subsystem regression parameters of Hammerstein model. To prove the effectiveness of this method, the comparative experiments with the traditional identification method are carried out, and this method is applied to modeling and control of the fractionating system in a petroleum refinery. The experimental results indicates that the proposed method guarantees the modeling accuracy of the fractionating system, and improves the tracking control performance.