Enhanced cryogenic distillation column identification for methane separation: a hybrid artificial neural network approach

Abstract Modelling the dynamics of cryogenic distillation columns is challenging due to their complex, nonlinear behaviour. This study introduces a novel identification approach using a hybrid Artificial Neural Network (ANN) optimized with Particle Swarm Optimization (PSO), applied to cryogenic distillation as a case study. The NARX-PSO-ANN model effectively captures the nonlinear dynamics of the distillation process by optimizing model parameters and avoiding local optima. The novelty of this work lies in integrating the NARX (Nonlinear Autoregressive with Exogenous Inputs) architecture with PSO, which enhances robustness and performance. To validate the model’s efficacy, realistic simulations of the cryogenic distillation column were conducted using Aspen Plus Dynamics, generated 2,000 data samples-1,400 training and 600 for validation. The NARX-PSO-ANN model was evaluated against established methods like BP-ANN and NARX-based BP-ANN, consistently outperforming them in identifying cryogenic distillation column dynamics and demonstrating superior effectiveness for complex separation processes. A user-friendly Python-based graphical user interface (GUI) was developed for real-time methane composition prediction, making the model accessible for practical applications. This innovative approach offers a reliable solution for optimizing complex, nonlinear systems in the process industry.