Review and Comparison Between Different Surrogate Models for Analysis of Catalytic Fixed-bed Reactor
Volume Title: 1
1Chemical Engineering, Engineering, Ferdowsi University of Mashhad, Mashhad, Iran
2Research Institute of Petroleum Industry(RIPI)
The need for accurate, fast and low-cost methods to predict characteristics of complex chemical reactions has led to the use of various models in fields of science and engineering. Chemical engineering and its subcategories are not excluded from this rule. In chemical engineering, surrogate models are used for various purposes, including modeling, optimization and, etc. Latin Hypercube is used as a promised sampling method since it has better performance compared to other sampling techniques based on our previous study. In the current study different types of alternative models: Linear Regression, Support Vector Regression (SVR), Multi-layer Perceptron Neural Network (MLP), Radial Neural Network are studied and compared based on error rates. Results indicated that MLP provides the best ability to predict the steady-state behavior temperature, pressure, mole fraction of components, maximum temperature of a direct dimethyl ether synthesis in the fixed-bed reactor.