Neural-Network Simulation of Catalytic Processes in an Adsorption Layer

Abstract

Two mutually complementary approaches are commonly used in the mathematical simulation of processes in heterogeneous catalysis. The first approach, which is called deterministic, is based on the solution of chemical kinetics differential equations for processes that occur in the system [1‐4]. The other approach, which is called statistical, is implemented using the Monte Carlo method [5] for the lattice gas model on rather arbitrary assumptions on the character of the interactions in the adsorption layer. This approach makes it possible to construct a wide range of models capable of reproducing complex nonlinear effects such as surface phase transitions, propagation of antiwaves at the surface of a catalyst, etc. [6‐11]. One of the disadvantages of this approach consists in the absence of suitable methods for qualitative analysis [12]. In addition, the problem of the consistent macro- and microlevel description of processes at the surface of a catalyst remains unsolved. In this paper, we first proposed a method for changing from the representation of a system as a statistical model to the deterministic description. Within this method, a series of statistical models are constructed, and their output data are processed using neural-network techniques. It was demonstrated that the neuralnetwork simulation of catalytic processes in an adsorption layer makes it possible to relate characteristics of the micro- and macrolevel description of a system. In the majority of cases, the general scheme of the statistical simulation of processes in heterogeneous catalysis by the Monte Carlo method is as follows [8]: