Search

Hydrogen (H2) is an environmentally-safe power source and its demands is continuously growing worldwide. The most important approach for its generation is water–gas shift (WGS) reaction through various catalysts. This work investigates feasibility of neural network method named Multilayer Perceptron Neural Network (MLP-NN) to estimate CO conversion in WGS reactions based on different active phase compositions and various supports. The approach considers the intrinsic parameters of the catalyst to estimate reaction performance. This research investigates the most influential variables by conducting a sensitivity analysis study on the predictions of the implemented method. The results of the modeling study revealed that the MLP-NN method can accurately approximate the experimental CO conversion values. The sensitivity analysis study revealed temperature and H2 feed concentration are the most crucial parameters on the reaction performance. The reliability of neural network methods is proved such as the MLP-NN to accurately estimate the CO conversion values in WGS reaction. [ABSTRACT FROM AUTHOR]

Hydrogen (H 2) is known as an environmentally friendly and crucial source of energy and recently demands for this compound has increased globally. One of the main ways of production of this compound is the Water-Gas Shift (WGS) reaction which utilizes various catalysts based on the operating conditions of the process. This work presents the performance of a soft computing method named PSO-RBF to estimate carbon monoxide (CO) conversion in WGS reactions. In addition, decision tree analysis was also performed to extract the associated rules which provides the highest performance for the reaction (higher conversion of CO). Different active phase compositions and support types of catalysts were utilized for development of the PSO-RBF model. The developed model accounts for the intrinsic catalyst parameters for estimation of the reaction performance by including features such as surface area, calcination time and temperature. In addition, sensitivity analysis of the model predictions was also examined to identify useful patterns. The results showed that the PSO-RBF model can accurately predict the actual CO conversion data with overall R2, AARD%, and RMSE values of 0.9977, 3.93, and 0.0159, respectively. It was also observed that the decision tree model can successfully extract the rules and trends from the experimental data. The outcomes of the sensitivity analysis study revealed that the most influential parameters for the process are temperature and H 2 composition in the feed stream. This work shows the capability of soft computing methods such as the PSO-RBF and decision tree approaches for estimating better catalysts and process conditions for this crucial reaction in the environmental field. • A PSO-RBF method for hydrogen production prediction for the Water-Gas Shift reaction was proposed. • Sensitivity analysis was performed to investigate the changes in conversion of CO. • Decision tree analysis was performed to find best catalyst/reaction conditions. • The approach exhibits accurate results for different catalysts and operating conditions during the process. • The temperature and the H 2 content of the feed stream are the most significant parameters. [ABSTRACT FROM AUTHOR]

Abstract Hydrogen (H 2) is considered a clean valuable energy source and its worldwide demand has increased in recent years. The Water-Gas Shift (WGS) Reaction is one of the major routes for hydrogen production and uses different catalysts depending on the operating process conditions. A catalyst is usually composed of an active phase and a support for its dispersion. There are currently an increasing number of researches on catalytic field focusing on transition metals nanoparticles supported on different compounds. In order to predict optimal catalyst compositions for the WGS reaction, Artificial Neural Networks (ANNs) were used to build a model from the literature catalytic data. A three-layer feedforward neural network was employed with active phase composition and support type as some of the input variables, and Carbon Monoxide (CO) conversion as output variable. The insertion of properties such as surface area, calcination temperature and time allowed predicting the reaction performance based on intrinsic catalyst variables not commonly used in phenomenological kinetic models. Also, unlike previous studies, a detailed sensitivity analysis was carried out to observe useful trends. An important outcome of this work is the proposition of ceria-supported catalysts for the WGS reaction that present larger surface areas, with Ru, Ni or Cu as active phases conducted at moderate temperatures (≈300 °C) and with reasonable space velocities (2000–6000 h−1). In addition, it was possible to predict the most relevant variables for the process: the temperature and the surface area. Thus, the results show the power of ANNs for predicting better catalysts and conditions for this important process in the environmental field. Graphical abstract Image 1 Highlights • An ANN model for catalyst selection for the Water-Gas Shift reaction was developed. • Changes in CO conversion were investigated by performing a sensitivity analysis. • The model can predict better catalysts and operating conditions for the process. • The temperature and the surface area are the most significant variables. [ABSTRACT FROM AUTHOR]