ANN-Based Surrogate Modeling for Prediction and Optimization of Carbon Conversion to Methanol Plant Under Uncertainty

Abstract

Artificial Neural Networks (ANN) utilization as surrogates within particle swarm optimization (PSO) and genetic algorithm (GA) frameworks for methanol flow rate optimization under uncertainty is explored in this work. First, Aspen Plus model with steady-state conditions of the CO2 hydrogenation to methanol process was developed. The process model was then transformed to a dynamic mode by inserting ±5% uncertainty in the process and 3880 data samples were generated. An ANN model, developed using MATLAB 2022a, was trained using these data samples. ANN achieved an impressive accuracy having a root mean square error (RMSE) of 26.83 and correlation coefficient (R) of 0.995 while testing for unseen data during cross-validation of the model. Then the ML model ANN was used as a surrogate in the PSO and GA for optimization methods to identify optimal conditions that maximize the methanol flow rate amidst uncertainty. Results show consistent improvements over the standalone Aspen model, with PSO slightly outperforming GA. Validation in Aspen Plus confirms the efficacy of the proposed methodology. This study highlights the potential of ANN-based surrogate modeling and its application in intelligent data-driven optimization for complex chemical processes under uncertainty, ultimately contributing to more efficient production systems.