Comparison of different sampling and surrogate modelling approaches for a multi-objective optimization problem of direct dimethyl ether synthesis in the fixed-bed reactor

It is a difficult task to select one proxy model with the proper range of accuracy and speed among various models. In the current study various types of sampling and surrogate models of a catalytic fixed-bed reactor (direct synthesis of DME) have been compared to perform a multi-objective optimization problem based on data. The challenge is to present a proper model which is capable of introducing a group of optimum points (as the Pareto front) in a single run in a multi-objective optimization problem based on Genetic Algorithm (GA) with comparable accuracy to the exact model but lower computational time during optimization. The initial database has been produced by different sampling methods. By applying the proposed method, the Pareto Front using surrogate models is achieved faster than the exact model, with quite satisfying accuracy. Different types of errors have been studied in sampling methods such as Random, Latin Hypercube Sampling (LHS), Sobol, and Halton, as well as alternative models, Linear Regression, Multilayer Perceptron Neural Network (MLP), Radial Basis Function Neural Network, Support Vector Regression, and Automatic Learning of Algebraic Models (ALAMO). Obtained results indicated that LHS as the sampling method and MLP and ALAMO as the surrogate models have the best ability to predict and optimize the steady-state behavior of the direct dimethyl ether synthesis in the fixed-bed reactor with the lowest mean square error (MSE) of 0.00027. •Modeling of Dimethyl-Ether synthesis from syngas in a catalytic Fixed-bed Reactor as a highly nonlinear case study.•Investigating different types of sampling and surrogate models.•Selecting LHS as the best sampling method and MLP and ALAMO as the proper alternative models with MSE of 0.000268•Efficient Multi-objective optimization based on the surrogate model.•Obtaining trade-off solutions with proper accuracy based on pareto front using surrogate model.