基于模型参数实时更新的FCC产品产率预测混合建模  

Hybrid Prediction Model for FCC Product Yields Based on Real-Time Model Parameter Updating

作　　者：薛腾方 周华[1] 魏彬 XUE Tengfang;ZHOU Hua;WEI Bin(Department of Chemical and Biochemical Engineering,College of Chemistry and Chemical Engineering,Xiamen University,Xiamen 361005,China)

机构地区：[1]厦门大学化学化工学院化学工程与生物工程系,福建厦门361005

出　　处：《石油学报(石油加工)》2025年第2期332-343,共12页Acta Petrolei Sinica(Petroleum Processing Section)

基　　金：国家自然科学基金项目(22272138)资助。

摘　　要：针对流化催化裂化(FCC)集总反应动力学模型参数难以适应反应过程中工况和原料变化等情况,基于代理模型所获取的模拟数据建立预测FCC产品产率的混合模型。该模型利用机器学习实时更新集总动力学模型参数,避免了纯数据驱动模型可解释性较差问题,弥补了集总动力学模型参数适配难的不足。结果表明,该混合模型具有较高的预测精度,其预测主要产品柴油、汽油和液化气的产率平均绝对百分比误差(MAPE)均小于2%。此外,在FCC原料性质波动、装置中关键参数检测信号含有噪声时,混合模型依旧保持较好的预测性能,其预测主要产品柴油、汽油和液化气的产率MAPE仍保持小于2%。由此说明所建立的混合模型具有较好的泛化性与鲁棒性,可为工业装置设计及操作优化提供理论指导。In view of the problem that it is difficult for parameters of the lumped kinetics model to match the varying working conditions and feedstock changes in the fluid catalytic cracking(FCC)process,a hybrid model for predicting FCC product yields has been established based on simulation data obtained from the agent model,which updates the model parameters of lumped kinetics in real time based on machine learning.This model can solve the problem of poor interpretability in data-driven models and the difficulty of matching between lumped kinetic model parameters and working conditions.The results show that the hybrid model has high prediction accuracy,with a mean absolute percentage error(MAPE)of the main products(diesel,gasoline,and liquefied petroleum gas)yields below 2%.In addition,the hybrid model still maintains a desirable predictive performance when the FCC feedstock properties fluctuate and the detection signals of key parameters in the unit contain noises,with a MAPE of the main products(diesel,gasoline,and liquefied petroleum gas)yields still below 2%.It thus demonstrates that the hybrid model has favorable generalization and robustness,which means it can provide theoretical guidance for design and operation optimization.

关 键 词：流化催化裂化 产率预测 混合模型 集总动力学 机器学习 

分 类 号：TE624[石油与天然气工程—油气加工工程]