应用经遗传算法优化的BP神经网络预测催化裂化装置焦炭产率  

Predicting coke yield of FCC unit using genetic algorithm optimized BP neural network

作　　者：苏鑫[1] 裴华健 吴迎亚 高金森[1] 蓝兴英[1] 

机构地区：[1]中国石油大学(北京)重质油国家重点实验室,北京102249

出　　处：《化工进展》2016年第2期389-396,共8页Chemical Industry and Engineering Progress

摘　　要：焦炭是催化裂化装置的主要副产物,准确预测催化裂化焦炭产率对提高装置的操作平稳度和经济效益具有重要意义。人工神经网络(ANN)具有强大的自学习和自适应能力,在非线性预测方面具有明显的优势。本研究将遗传算法(GA)与BP神经网络相结合,基于某炼厂催化裂化装置的生产数据,分别从原料、催化剂和操作条件3个方面选取28个关键影响参数建立了催化裂化焦炭产率预测模型,分别将BP神经网络和经遗传算法优化的BP神经网络(GA-BP)的预测结果与工业数据进行对比。结果表明,经遗传算法优化的预测模型无论在预测结果的准确性还是稳定性方面效果更好。最后,本研究还通过考察原料残炭、反应温度等单一关键参数对焦炭产率的影响,进一步证明了经遗传算法优化的BP神经网络预测模型的准确性。Coke is the main by-product of fluid catalytic cracking（FCC） process. It is of great significance to predict coke yield accurately to enhance stability and economic performance of FCC plant. Artificial neural network（ANN） has a strong self-learning and adaptive ability,and has obvious advantages in nonlinear forecasting. In this paper,a new model combining BP neural network and genetic algorithm（GA） was developed to predict coke yield by choosing 28 key parameters involving feedstock properties,catalyst properties and operating conditions of industrial data of FCC unit,The prediction results obtained from BP neural network and the genetic algorithm optimized BP neural network（GA-BP） were compared. The GA-BP model had a better result in both accuracy and stability. Furthermore,the influence of key parameters,such as reaction temperature,feedstock carbon residue on coke yield was investigated,which further proved the accuracy of BP neural network model optimized by genetic algorithm.

关 键 词：催化裂化 焦炭产率 神经网络 遗传算法 

分 类 号：TE624[石油与天然气工程—油气加工工程]