Cost-optimized design of front-end acetylene hydrogenation using artificial neural networks

Given that acetylene is a poison to ethylene polymerization catalysts, this research studied the design for a cost-effective optimization of the front-end acetylene hydrogenation unit. The design capacity is 0.6 million ton/annum (for a commercial petrochemical plant) to reduce the acetylene concentration to five ppm while maintaining the unit’s controllability. The study started by modeling an adiabatic fixed-bed pseudo-homogeneous reactor with Pd-Ag/Al₂O₃ catalyst. The hydrogenation plant was designed using the validated model and literature. It was also possible to do sensitivity analysis on the important operational parameters, like temperature, pressure, the number of catalytic beds, and volumes, in order to get 12467 samples ready for training the artificial neural network (ANN) model. In the next step, the economic potential of the artificial neural network model was optimized with the genetic algorithm (GA). In this optimization problem, economic potential was the fitness function, and utility consumption in heat exchangers, feed pressure, bed volumes, and presence of the third bed were the optimization variables. Optimization using GA-ANN model, GA for numerical model, and Bayesian for ANN revealed that the GA-ANN approach significantly reduced function evaluations (27,900 vs. 51,900 in GA-Numerical) and achieved the highest economic potential (1.997 million USD annually). The optimization led to a reduction in normal operating temperature and inlet pressure.