Multiobjective Optimization of CO2 Emission and Net Profit for a Naphtha Cracking Furnace Using a Deep Neural Network with a Nondominated Sorting Genetic Algorithm

A naphtha-cracking furnace converts naphtha to ethylene (EL) and propylene (PL); the yields depend on the coil outlet temperature (COT) and naphtha composition. However, determining the optimal COT for maximizing net profit is difficult because the product price and its composition fluctuate frequently. Moreover, CO2 emissions increase inevitably with increasing net profit, which requires taking environmental aspects into account. Hence, this study proposes a multiobjective optimization model for the naphtha cracking furnace by considering the incompatible goals: maximization of net profit and minimization of CO2 emissions. First, a deep neural network (DNN)-based model is developed to predict the EL yield, PL yield, and CO2 emissions for a given COT and naphtha composition using 783 industrial data points. Second, the developed model is combined with a nondominated sorting genetic algorithm (NSGA-II) for multiobjective optimization to obtain a Pareto front with various solutions. Finally, case studies are conducted for different product prices: EL was more expensive than PL in 2018; PL was more expensive than EL in 2019; and EL and PL had similar prices in 2020. For these three cases, the actual industrial data are applied to the model, and various solutions are proposed. The representative solutions in each case exhibit 5.35-6.14% higher net profits and 12.81-15.34% lower CO2 emissions than those of the industrial data. The proposed model can help decision-makers by providing flexible options for the modification of various production parameters, including environmental regulations.