Pouya Es'haghi; Keivan Shayesteh; Hassan Seddighi
Abstract
In chemical industries, precision in calculations and process simulations is crucial. One of the most influential parameters is the molar density of a fluid under various pressure and temperature conditions. Equations of state (EOS) are common among the methods for determining molar density. Usually, ...
Read More
In chemical industries, precision in calculations and process simulations is crucial. One of the most influential parameters is the molar density of a fluid under various pressure and temperature conditions. Equations of state (EOS) are common among the methods for determining molar density. Usually, the error resulting from predicting molar density using EOS is generally high at high temperatures and pressures due to the increased intermolecular effects. Additionally, due to the form of EOS concerning volume or molar density, calculating molar volume at specified temperature and pressure requires suitable numerical methods for root-finding. This article aims to present an effective method for estimating the molar density of benzene using two crucial machine learning methods, namely Multi-Layer Perceptron-Artificial Neural Network(MLP-ANN) and Adaptive Neuro-Fuzzy Inference System (ANFIS). This study used 302 sets of experimental data to train these two methods. Additionally, another set of 60 experimental data was used to compare the errors of the methods. The Peng-Robinson (PR) equation was also employed in this article to evaluate the performance of machine learning methods better and calculate molar density. The results showed that the mean relative errors (MRE) for the MLP-ANN, ANFIS, and PR methods for the 362 data points are 0.838%, 1.791%, and 4.834%, respectively. The results demonstrated that using machine learning methods can reduce computational errors, with the error from predicting using the PR equation being almost five times that of MLP-ANN. In this article, the MLP-ANN method outperformed ANFIS due to its computational efficiency and lower error in predicting molar density.
