Modeling, Sensitivity Analysis, and Optimization of the Methanol-to-Gasoline Process using Artificial Intelligence Methods

Abstract

In this study, the gasoline yield in the methanol-to-gasoline (MTG) process was modeled using artificial neural network (ANN) and multivariate polynomial regression (MPR) techniques. The ANN trained using the Levenberg–Marquardt (LM) method and having three neurons in the hidden layer was the most accurate at predicting gasoline yield (R² = 0.993 and RMSE = 0.024). Therefore, this network was used to investigate the influence of operational conditions such as pressure, weight hourly space velocity (WHSV), temperature, and the average particle size of the Zeolite Socony Mobil–5 (ZSM-5) catalyst on the gasoline yield. Then, the particle swarm optimization (PSO) and genetic algorithm (GA) were used to approach the best operating parameters and catalyst size to get the most gasoline yield. The mentioned neural network was used as a fitness function in the optimization algorithms. The optimization results showed that at a pressure of 1 bar, a temperature of 400°C, a WHSV equal to 1 h^–1, and a particle size of 1466 nm, the maximum gasoline yield is equivalent to 45.43.