Nonlinear Modelling for Steam Temperature of Distillation Column: A Comparison between PRBS and MPRS Perturbation Signals

Abstract

Nowadays, countless research efforts have been reported for distillation column on nonlinear modelling. Several studies have shown the importance of appropriate perturbation signal for the nonlinear system applications. This study focuses on a nonlinear modelling for steam temperature using Pseudo Random Binary Signal (PRBS) and Multi-level Pseudo Random Sequence (MPRS) perturbation signals. A Binary Particle Swarm Optimisation (BPSO) algorithm was utilised in the model structure selection for polynomial Nonlinear Auto-Regressive with eXogenous (NARX) input for Steam Distillation Pilot Plant (SDPP). Three model’s selection criteria were examined; Akaike Information Criterion (AIC), Model Descriptor Length (MDL), and Final Prediction Error (FPE). The performance analysis included output model analysis and model validation of the nonlinear model. The results demonstrated fewer number of input and output lags and lesser amount of output model parameter using MPRS perturbation signal compared with PRBS. Model validation showed high R-squared and low MSE for both signals’ application.