A comparative study of thermodynamic equations and artificial neural networks in modeling the behavior of glycerol+methanol+CO2 and glycerol+ethanol+CO2 systems in biodiesel production

• Published in: Chemical engineering research & design 2024-02, Vol.202, p.92-102
• Main Authors: Klauck, Gabriel, Dalmolin, Irede, Zanini Brusamarello, Claiton
• Format: Article
• Language: English
• Subjects: Elman Network; Feed-Forward Network; Glycerol; Thermodynamic
• ISSN: 0263-8762
• DOI: 10.1016/j.cherd.2023.12.013

Classical thermodynamic methods often encounter challenges when applied to model the intricate behavior of liquid-liquid equilibria within biofuel production systems. To overcome these limitations, Artificial Neural Networks (ANNs) have emerged as a promising avenue for effectively describing such complex systems. This study developed and evaluated ANNs to predict the behavior of two key systems, glycerol+ethanol+CO2 and glycerol+methanol+CO2, employing diverse configurations and parameters. A total of 516 distinct ANNs were methodically crafted and evaluated. Notably, the Cascade Feed-Forward architecture exhibited optimal performance in predicting the CO2+glycerol+ethanol system, while the Elman networks excelled in predicting the CO2+glycerol+methanol system. The optimized ANNs were subsequently benchmarked against analogous systems in the literature by applying established thermodynamic equations. Remarkably, the results of this research work underscore the superior predictive accuracy of the ANN approach, yielding significantly diminished error values. Specifically, the mean squared error (MSE) values of 0.1283 and 0.0892 were achieved for the methanol and ethanol systems, respectively. This study contributes substantively by showcasing the ability of ANNs to surpass classical methods for accurate prediction within the intricate domain of biofuel production systems. [Display omitted] •Optimal ANNs predict CO2+ glycerol systems accurately (MSE: 0.13; 0.09).•Predict results with a relatively small number of iterations.•ANNs offer efficiency and versatility in complex system predictions.