基于无监督时序聚类的催化裂化装置工况划分识别与产率预测方法

doi:10.11949/0438-1157.20241453

摘要/Abstract

摘要：

作为炼厂中的核心生产装置，催化裂化装置将重质油转化为轻质油品，是提升炼厂经济效益的重要环节。由于生产工艺的复杂性以及我国石化行业原油的频繁变化，基于流程模拟的催化裂化装置模型预测精度难以满足实时优化的需求。为此，本文提出了一种基于无监督时序聚类的催化裂化装置工况划分识别与产率预测方法，通过挖掘过程数据中的有益信息，实现工况的自动划分与识别，提高了在多原油混炼条件下的产率预测性能。通过对实际数据的应用分析，验证了所提方法具有良好的预测能力和泛化能力，能够实现催化裂化装置产品产率的高精度实时预测，从而有效地满足催化裂化装置动态实时优化的需求，有利于提升炼厂的经济效益。

关键词: 催化裂化装置, TICC聚类, 产率预测, 数据驱动建模, 过程控制

Abstract:

As a core production unit in the refinery, the catalytic cracking unit converts heavy oil into light oil products, which is an important link in improving the economic benefits of the refinery. Due to the complexity of the production process and the frequent variation in crude oil types in petrochemical industries of China, the prediction accuracy of catalytic cracking unit models based on process simulation often fails to meet the requirements of real-time optimization. To address this, a novel method for operating conditions classification and yield prediction based on unsupervised time series clustering is proposed in this study. By extracting valuable information from process data, the method achieved automatic classification and identification of operating conditions, thereby improving yield prediction performance under conditions of mixed crude oil processing. Through practical data analysis, the proposed method was demonstrated to possess strong predictive capability and generalization ability, enabling high-precision real-time yield prediction for catalytic cracking unit products. This approach can effectively meet the need for dynamic real-time optimization of catalytic cracking units, thereby benefiting the promotion of economic performance in refineries.

Key words: FCCU, TICC clustering algorithm, yield prediction, data-driven modeling, process control

中图分类号:

TP 274

张涵川, 尚超, 吕文祥, 黄德先, 张亚宁. 基于无监督时序聚类的催化裂化装置工况划分识别与产率预测方法[J]. 化工学报, 2025, 76(6): 2781-2790.

Hanchuan ZHANG, Chao SHANG, Wenxiang LYU, Dexiang HUANG, Yaning ZHANG. Operating conditions pattern recognition and yield prediction for FCCU based on unsupervised time series clustering[J]. CIESC Journal, 2025, 76(6): 2781-2790.

图/表 8

图1 基于无监督时序聚类的催化裂化装置工况划分识别方法流程

Fig.1 Pipeline of classification and identification for FCCU operating conditions based on unsupervised time series clustering

图2 TICC聚类算法流程

Fig.2 TICC clustering algorithm process

图3 基于多模型的催化裂化装置在线实时产率预测方法流程

Fig.3 Process of online real-time yield prediction method for FCCU based on multiple models

图4 基于催化裂化装置在线工况识别的产率预测框架应用

Fig.4 Application process of yield prediction framework for identification of online operating conditions of FCCU

图5 聚类结果(时间)

Fig.5 Clustering results(time)

图6 聚类结果(数量)

Fig.6 Clustering results (quantity)

表1 MPM与OPM预测MAPE对比（总体数据集）

Table 1 Comparison of prediction MAPE between MPM and OPM（overall dataset）

模型	汽油产率预测MAPE		柴油产率预测MAPE
模型	总训练集	总测试集	总训练集	总测试集
OPM	7.59%	7.66%	4.72%	4.71%
MPM	2.89%	2.98%	2.68%	2.73%

表2 MPM与OPM预测MAPE对比（分类数据集）

Table 2 Comparison of prediction MAPE between MPM and OPM（classification dataset）

模型	汽油产率预测MAPE		柴油产率预测MAPE
模型	训练集1	测试集1	训练集1	测试集1
OPM	7.84%	7.86%	3.39%	3.35%
MPM₁	4.11%	4.20%	2.73%	2.77%
	训练集2	测试集2	训练集2	测试集2
OPM	4.24%	4.20%	5.08%	5.35%
MPM₂	1.87%	1.92%	3.74%	4.01%
	训练集3	测试集3	训练集3	测试集3
OPM	8.46%	8.18%	15.30%	15.41%
MPM₃	1.35%	1.36%	2.33%	2.52%
	训练集4	测试集4	训练集4	测试集4
OPM	4.27%	4.46%	3.59%	3.42%
MPM₄	2.37%	2.43%	2.42%	2.20%
	训练集5	测试集5	训练集5	测试集5
OPM	8.09%	8.57%	5.52%	5.31%
MPM₅	3.80%	3.57%	3.74%	3.68%
	训练集6	测试集6	训练集6	测试集6
OPM	14.05%	14.86%	6.56%	6.55%
MPM₆	3.41%	3.68%	1.66%	1.68%
	训练集7	测试集7	训练集7	测试集7
OPM	8.99%	9.11%	4.37%	4.22%
MPM₇	1.14%	1.01%	1.85%	1.77%
	训练集8	测试集8	训练集8	测试集8
OPM	7.29%	7.15%	4.92%	4.90%
MPM₈	3.02%	3.35%	2.48%	2.54%

参考文献 30

[1]	郑远扬, 高少立, 袁璞. 催化裂化装置的动态模型: Ⅰ. 提升管反应器的动态模型和动力学参数的估计[J]. 石油炼制与化工, 1986, 17(2): 23-30.
	Zheng Y Y, Gao S L, Yuan P. Dynamic model of catalytic cracking unit: Ⅰ. Dynamic model of riser reactor and estimation of kinetic parameters[J]. Petroleum Processing and Petrochemicals, 1986, 17(2): 23-30.
[2]	郑远扬, 高少立. 催化裂化装置的动态模型(Ⅱ)提升管反应器的集中参数模型[J]. 石油炼制与化工, 1986, 17(4): 67-71.
	Zheng Y Y, Gao S L. Dynamic model of catalytic cracking unit (Ⅱ) lumped parameter model of riser reactor[J]. Petroleum Processing and Petrochemicals, 1986, 17(4): 67-71.
[3]	郑远扬, 高少立. 催化裂化装置的动态模型: Ⅲ. 两段再生器的动态模型和动力学参数估计[J]. 石油炼制与化工, 1986, 17(5): 45-49.
	Zheng Y Y, Gao S L. Dynamic model of catalytic cracking unit Ⅲ. Dynamic model of two-stage regenerator and estimation of dynamic parameters[J]. Petroleum Processing and Petrochemicals, 1986, 17(5): 45-49.
[4]	Zheng Y Y. Dynamic modeling and simulation of a catalytic cracking unit[J]. Computers & Chemical Engineering, 1994, 18(1): 39-44.
[5]	陈玉石. FCCU反应再生部分动态建模与仿真系统研究[D]. 厦门: 厦门大学, 2007.
	Chen Y S. Research on dynamic modeling and simulation system of FCCU reaction regeneration part[D]. Xiamen: Xiamen University, 2007.
[6]	邵帅. R2R型催化裂化装置反-再系统的建模与动态仿真[D]. 北京: 北京化工大学, 2008.
	Shao S. Modeling and dynamic simulation of reaction-regeneration system in R2R catalytic cracking unit[D]. Beijing: Beijing University of Chemical Technology, 2008.
[7]	Radu S, Ciuparu D. Modelling and simulation of an industrial fluid catalytic cracking unit[J]. Revista de Chimie, 2014, 65(1): 113-119.
[8]	Singh B, Sahu S, Dimri N, et al. Seventeen-lump model for the simulation of an industrial fluid catalytic cracking unit (FCCU)[J]. Sādhanā, 2017, 42(11): 1965-1978.
[9]	Orazbayev B, Kozhakhmetova D, Wójtowicz R, et al. Modeling of a catalytic cracking in the gasoline production installation with a fuzzy environment[J]. Energies, 2020, 13(18): 4736.
[10]	Ni P, Liu B, He G. An online optimization strategy for a fluid catalytic cracking process using a case-based reasoning method based on big data technology[J]. RSC Advances, 2021, 11(46): 28557-28564.
[11]	Li T Y, Long J, Zhao L, et al. A bilevel data-driven framework for robust optimization under uncertainty-applied to fluid catalytic cracking unit[J]. Computers & Chemical Engineering, 2022, 166: 107989.
[12]	Hong J, Tian W D. Prediction in catalytic cracking process based on swarm intelligence algorithm optimization of LSTM[J]. Processes, 2023, 11(5): 1454.
[13]	Liu N, Zhu C M, Zhang M X, et al. A multiscale adaptive framework based on convolutional neural network: application to fluid catalytic cracking product yield prediction[J]. Petroleum Science, 2024, 21(4): 2849-2869.
[14]	Zhou C, Liu Q Y, Huang D X, et al. Inferential estimation of kerosene dry point in refineries with varying crudes[J]. Journal of Process Control, 2012, 22(6): 1122-1126.
[15]	Gao X Y, Shang C, Jiang Y H, et al. Refinery scheduling with varying crude: a deep belief network classification and multimodel approach[J]. AIChE Journal, 2014, 60(7): 2525-2532.
[16]	Das S, Nayak M, Senapati M R. Improving time series prediction with deep belief network[J]. Journal of the Institution of Engineers (India): Series B, 2023, 104(5): 1103-1118.
[17]	Hallac D, Vare S, Boyd S, et al. Toeplitz inverse covariance-based clustering of multivariate time series data[C]//Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, August 13-17, 2017, Halifax, NS, Canada. ACM, 2017: 215-223.
[18]	Li H H, Liu J X, Yang Z L, et al. Adaptively constrained dynamic time warping for time series classification and clustering[J]. Information Sciences, 2020, 534: 97-116.
[19]	Ienco D, Interdonato R. Deep multivariate time series embedding clustering via attentive-gated autoencoder[C]//PAKDD 2020, May 11-14, 2020, Singapore, Springer International Publishing, 2020: 318-329.
[20]	Huang H, Shah T, Yoo S. Deep time series sketching and its application on industrial time series clustering[C]//2022 IEEE International Conference on Big Data (Big Data), December 17-20, 2022, Osaka, Japan. IEEE, 2022: 1997-2006.
[21]	Ienco D, Interdonato R. Deep semi-supervised clustering for multi-variate time-series[J]. Neurocomputing, 2023, 516: 36-47.
[22]	Ismail Fawaz H, Forestier G, Weber J, et al. Deep learning for time series classification: a review[J]. Data Mining and Knowledge Discovery, 2019, 33(4): 917-963.
[23]	徐春明, 杨朝合. 石油炼制工程[M]. 4版. 北京: 石油工业出版社, 2009.
	Xu C M, Yang C H. Petroleum Refining Engineering[M]. 4th ed. Beijing: Petroleum Industry Press, 2009.
[24]	Koller D, Friedman N. Probabilistic Graphical Models: Principles and Techniques[M]. Cambridge, Massachusetts: The MIT Press, 2009.
[25]	Lauritzen S L. Graphical Models[M]. Oxford: Clarendon Press, 1996.
[26]	Friedman J, Hastie T, Tibshirani R. Sparse inverse covariance estimation with the graphical lasso[J]. Biostatistics, 2008, 9(3): 432-441.
[27]	Stephen B, Parikh Neal, Eric C, et al. Distributed optimization and statistical learning via the alternating direction method of multipliers[J]. Foundations and Trends^® in Machine Learning, 2011, 3(1):1-122.
[28]	张学工. 关于统计学习理论与支持向量机[J]. 自动化学报, 2000, 26(1): 32-42.
	Zhang X G. Introduction to statistical learning theory and support vector machines[J]. Acta Automatica Sinica, 2000, 26(1): 32-42.
[29]	Hsu C W, Lin C J. A comparison of methods for multiclass support vector machines[J]. IEEE Transactions on Neural Networks, 2002, 13(2): 415-425.
[30]	Gao T Y, Luo H, Yin S, et al. A recursive modified partial least square aided data-driven predictive control with application to continuous stirred tank heater[J]. Journal of Process Control, 2020, 89: 108-118.