Neural Network-based Hybrid Estimator for Estimating Concentration in Ethylene Polymerization Process: An Applicable Approach

Abstract

Estimation of a monomer concentration of an ethylene polymerization process has been a challenging problem due to its highly nonlinear behavior and interaction among state variables.  Applying of an extended Kalman filter (EKF) to provide the estimates of the concentration based on measured bed temperatures has usually been prone to errors. Here, alternatively, neural network-based hybrid estimators have been developed and classified into three structures which integrating of either EKF or Kalman filter (KF) to neural network (NN) to provide the estimates. The NNs are integrated to provide the estimates’ error or concentration’s estimates corresponding to individual structure for reducing the estimation error. Simulation results have shown that the hybrid estimators can provide good estimates under nominal condition and disturbance cases. However, in dealing with noises, the NN-KF hybrid estimator gives superior robustness with smooth and accurate estimated values.