关键词: 催化裂化, 流程模拟, 深度神经网络, 多目标优化

Abstract: Due to the requirements of real-time optimization in the operation parameter optimization of the catalytic cracking reaction-regeneration system, and with the objective functions of maximizing the yield of ethylene plus propylene and minimizing the coke yield, a process mechanism model of the reaction-regeneration system was established using the advanced processed engineering expert process simulation software based on industrial data from a 3.0 Mt/a fluid catalytic cracking unit. Subsequently, a deep neural network model was adopted as the surrogate model, and multi-objective optimization of key parameters, including feed preheating temperature, reaction temperature, regeneration temperature, catalyst activity, and stripping steam flow rate, was performed using the non-dominated sorting genetic algorithm. The results show that the optimized combination of key parameters is as follows: preheating temperature 222.24 °C, reaction temperature 539.98 °C, regeneration temperature 719.98 °C, catalyst activity 58.13%, and stripping steam flow rate 11999.49 kg/h. Under the optimal operating conditions, the yield of ethylene plus propylene increases by approximately 1.96 percentage points, and the coke yield decreases by approximately 4.06 percentage points predicted by the model.

Key words: catalytic cracking, process simulation, deep neural network, multi-objective optimization