Control of the Gasoline Catalytic Reforming Process Based on a Kinetic Model Using an Artificial Neural Network

Abstract

This study presents an approach to accelerated modeling of one of the key processes in the oil refining industry — catalytic reforming of gasoline — using a deep neural network. The product of catalytic reforming constitutes the majority of commercial gasoline and determines its octane number. In addition, aromatic hydrocarbons are formed during the process, the content of which can be regulated by adjusting the reaction conditions. The aim of the study was to develop a model capable of replacing resource-intensive direct calculations for evaluating the output parameters of the process, thereby enabling real-time analysis and control. Traditional methods require significant time and computational resources, making them unsuitable for rapid deployment in industrial environments. The training dataset was generated based on the results of direct simulations performed in MATLAB. A significant reduction in data generation time was achieved through the use of parallel computing. Optimal hyperparameters for the neural network were identified. Model training was carried out using the minimization of the mean squared error (MSE) and the coefficient of determination (R²). The developed model demonstrates high prediction accuracy with minimal computational time, which makes it especially suitable for implementation in automated control systems at oil refineries. The resulting neural network approximation effectively acts as a simulator capable of instantaneously estimating key parameters of the catalytic reforming process, including reformate yield, content of aromatic hydrocarbons and benzene, and the octane number of the mixture.