Machine-learning-based optimization of operating conditions of naphtha cracking furnace to maximize plant profit

Abstract

Naphtha cracking is the primary process for propylene (PL) and ethylene (EL) production and depends on the operating conditions, such as the coil outlet temperature and feedstock composition. The product yields, in turn, determine the profit of the naphtha cracking plant. Therefore, the operating conditions should be optimized to maximize plant profit. However, it is challenging to optimize the conditions using conventional simulation methods since a high-fidelity model is hard to develop owing to the nonlinearity and complexity of the cracking process. In this study, we used machine learning to optimize the operating conditions of a naphtha cracking furnace to maximize plant profit. First, a data-driven model was developed to predict the product yields for different operating conditions using a deep neural network (DNN). The model could predict the PL and EL yields with high accuracy (R2 = 0.965 and 0.900, respectively). Next, a genetic algorithm was used for optimization based on the developed DNN model. Finally, the developed model was used with real-world plant data and product prices for 2020. The plant profit under the optimized operating conditions was 30% higher than that corresponding to the original operating conditions. Thus, the proposed method is suitable for determining the optimal operating conditions of various types of plants in order to maximize profit.