基于优化CNN-BiLSTM神经网络的间歇精馏过程建模

doi:10.11949/0438-1157.20250145

摘要/Abstract

摘要：

间歇式精馏因其操作灵活性和适应性广泛应用于精细化工、制药及食品加工等行业，然而其非稳态特性和显著变化的操作条件使得传统静态模型难以精确描述系统动态行为，最终导致物质在塔内的分离效果不佳。为此本研究以乙醇-水二元混合物体系间歇精馏塔馏出液和塔釜乙醇质量分数数据为研究对象，提出了一种基于卷积神经网络（convolutional neural network，CNN）和双向长短期记忆网络（bidirectional long short-term memory，BiLSTM）复合的预测软测量模型（CNN-BiLSTM），并通过改进的雪消融优化器（improved snow ablation optimizer，ISAO）优化模型超参数，旨在使其能够替代在线测量仪对间歇精馏控制起到辅助作用。实验结果表明，在馏出液和塔釜乙醇质量分数预测中，经过ISAO优化后的CNN-BiLSTM神经网络在测试集上的均方根误差和平均绝对误差相较于初始模型降幅至少为82.27%，其动态预测性能显著提升。

关键词: 间歇式, 二元混合物, 分离, 神经网络, 超参数优化, 优化算法

Abstract:

Batchwise distillation is widely employed in fine chemical, pharmaceutical, and food processing industries due to its operational flexibility and adaptability. However, its inherent non-steady-state characteristics and significantly varying operating conditions render traditional static models inadequate for accurately describing system dynamic behaviors, consequently leading to suboptimal separation efficiency of components within the column. To address this challenge, this study establishes a hybrid soft-sensor model (CNN-BiLSTM) integrating convolutional neural networks (CNN) and bidirectional long short-term memory networks (BiLSTM), specifically targeting the prediction of ethanol mass fractions in both distillate and bottom products of an ethanol-water binary mixture system. The model hyperparameters are systematically optimized through an improved snow ablation optimization (ISAO) algorithm, aiming to develop a reliable alternative to online measurement instruments for enhanced batch distillation control. The experimental results show that in the prediction of the mass fraction of distillate and bottom ethanol, the root mean square error and mean absolute error of the CNN-BiLSTM neural network after ISAO optimization on the test set are reduced by at least 82.27% compared with the initial model. This significant enhancement in dynamic prediction capabilities validates the proposed methodology's effectiveness in addressing the operational challenges inherent to batch distillation processes.

Key words: batchwise, binary mixture, separation, neural network, hyperparameter optimization, optimization algorithm

中图分类号:

O 6-31

郭旭, 贾继宁, 姚克俭. 基于优化CNN-BiLSTM神经网络的间歇精馏过程建模[J]. 化工学报, 2025, 76(9): 4613-4629.

Xu GUO, Jining JIA, Kejian YAO. Modeling of batch distillation process based on optimized CNN-BiLSTM neural network[J]. CIESC Journal, 2025, 76(9): 4613-4629.

图/表 22

参考文献 30

[1]	梁昕. 间歇精馏技术的优化综述[J]. 山东化工, 2016, 45(6): 41-42, 45.
	Liang X. Summary of batch distillation technology optimization[J]. Shandong Chemical Industry, 2016, 45(6): 41-42, 45.
[2]	Zavala-Loría J C, Narváez-García A. Batch Distillation: Thermodynamic Efficiency[M]. Rijeka: IntechOpen, 2012: 92-106.
[3]	杨海荣, 薄翠梅, 陆爱晶, 等. 间歇精馏过程变回流比智能控制策略的研究[J]. 化工自动化及仪表, 2008, 35(1): 12-16.
	Yang H R, Bo C M, Lu A J, et al. Study on variable reflux ratio intelligent control strategy for batch distillation process[J]. Control and Instruments in Chemical Industry, 2008, 35(1): 12-16.
[4]	Karacan S. Application of a non-linear long range predictive control to a packed distillation column[J]. Chemical Engineering and Processing: Process Intensification, 2003, 42(12): 943-953.
[5]	Henson M A. Nonlinear model predictive control: current status and future directions[J]. Computers & Chemical Engineering, 1998, 23(2): 187-202.
[6]	孙志军, 薛磊, 许阳明, 等. 深度学习研究综述[J]. 计算机应用研究, 2012, 29(8): 2806-2810.
	Sun Z J, Xue L, Xu Y M, et al. Overview of deep learning[J]. Application Research of Computers, 2012, 29(8): 2806-2810.
[7]	Karacan S, Hapoǧlu H, Alpbaz M. Application of optimal adaptive generalized predictive control to a packed distillation column[J]. Chemical Engineering Journal, 2001, 84(3): 389-396.
[8]	Alpbaz M, Karacan S, Cabbar Y, et al. Application of model predictive control and dynamic analysis to a pilot distillation column and experimental verification[J]. Chemical Engineering Journal, 2002, 88(1/2/3): 163-174.
[9]	陈锐, 贾继宁, 姚克俭. 基于优化BP神经网络的非稳态精馏过程建模[J]. 化学工程, 2023, 51(10): 83-88.
	Chen R, Jia J N, Yao K J. Modeling of unsteady distillation process based on optimized BP neural network[J]. Chemical Engineering (China), 2023, 51(10): 83-88.
[10]	Gu J X, Wang Z H, Kuen J, et al. Recent advances in convolutional neural networks[J]. Pattern Recognition, 2018, 77: 354-377.
[11]	Zrira N, Kamal-Idrissi A, Farssi R, et al. Time series prediction of sea surface temperature based on BiLSTM model with attention mechanism[J]. Journal of Sea Research, 2024, 198: 102472 .
[12]	Wright L G, Onodera T, Stein M M, et al. Deep physical neural networks trained with backpropagation[J]. Nature, 2022, 601(7894): 549-555.
[13]	Kavousi-Fard A, Samet H, Marzbani F. A new hybrid modified firefly algorithm and support vector regression model for accurate short term load forecasting[J]. Expert Systems with Applications, 2014, 41(13): 6047-6056.
[14]	Zhang Q, Wang H, Dong J Yet al. Prediction of sea surface temperature using long short-term memory[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(10): 1745-1749.
[15]	Deng L Y, Liu S Y. Snow ablation optimizer: a novel metaheuristic technique for numerical optimization and engineering design[J]. Expert Systems with Applications, 2023, 225: 120069.
[16]	Sun B, Li W, Huang Y. Performance of composite PPSO on single objective bound constrained numerical optimization problems of CEC 2022[C]//2022 IEEE Congress on Evolutionary Computation (CEC). Padua, Italy: IEEE, 2022: 1-8.
[17]	Hu G, Guo Y X, Wei G, et al. Genghis Khan shark optimizer: a novel nature-inspired algorithm for engineering optimization[J]. Advanced Engineering Informatics, 2023, 58: 102210.
[18]	董益斌, 熊敬超, 王敬宇, 等. 融合激光雷达料位测算的锅炉燃烧优化模型预测控制 [J]. 化工学报, 2024, 75(3): 924-935.
	Dong Y B, Xiong J C, Wang J Y, et al. LiDAR measurement based on model predictive control for boiler combustion optimization[J]. CIESC Journal, 2024, 75(3): 924-935.
[19]	Wang Z C, Xu N N, Bao X G, et al. Spatio-temporal deep learning model for accurate streamflow prediction with multi-source data fusion[J]. Environmental Modelling & Software, 2024, 178: 106091.
[20]	Schmidt-Hieber J. Nonparametric regression using deep neural networks with ReLU activation function[J]. The Annals of Statistics, 2020, 48(4): 1875-1897.
[21]	Elfwing S, Uchibe E, Doya K. Sigmoid-weighted linear units for neural network function approximation in reinforcement learning[J]. Neural Networks, 2018, 107: 3-11.
[22]	王贺琦, 王然, 刘诗琳, 等. 基于Tent映射的改进型人工蜂群算法[J]. 软件工程, 2022, 25(8): 15-19.
	Wang H Q, Wang R, Liu S L, et al. Modified artificial bee colony algorithm based on tent mapping[J]. Software Engineering, 2022, 25(8): 15-19.
[23]	刘三阳, 张晓伟. 混合差分变异策略[J]. 智能系统学报, 2008, 3(6): 487-491.
	Liu S Y, Zhang X W. A hybrid strategy for differential variation[J]. Caai Transactions on Intelligent Systems, 2008, 3(6): 487-491.
[24]	潘恒. 基于集成学习的个人信贷违约预测方法研究与应用[D]. 西安: 西安理工大学, 2024.
	Pan H. Research and application of personal credit default prediction method based on integrated learning[D]. Xi'an: Xi'an University of Technology, 2024.
[25]	Nadimi-Shahraki M H, Zamani H, Asghari Varzaneh Z, et al. A systematic review of the whale optimization algorithm: theoretical foundation, improvements, and hybridizations[J]. Archives of Computational Methods in Engineering, 2023, 30: 4113-4159.
[26]	Shehadeh H A. Chernobyl disaster optimizer (CDO): a novel meta-heuristic method for global optimization[J]. Neural Computing and Applications, 2023, 35(15): 10733-10749.
[27]	Trojovský P, Dehghani M. Subtraction-average-based optimizer: a new swarm-inspired metaheuristic algorithm for solving optimization problems[J]. Biomimetics, 2023, 8(2): 149.
[28]	Yue Y G, Cao L, Lu D W, et al. Review and empirical analysis of sparrow search algorithm[J]. Artificial Intelligence Review, 2023, 56(10): 10867-10919.
[29]	Wang D S, Tan D P, Liu L. Particle swarm optimization algorithm: an overview[J]. Soft Computing, 2018, 22(2): 387-408.
[30]	Xie J Y, He J C, Gao Z H, et al. An enhanced snow ablation optimizer for UAV swarm path planning and engineering design problems[J]. Heliyon, 2024, 10(18): e37819.

参数	数值
色谱柱温度/℃	180
进样口温度/℃	180
热导池温度/℃	200
桥电流/mA	110
氢气压力/MPa	0.01
进样体积/μl	1
色谱柱类型	Porapak-Q

参数	数值
色谱柱温度/℃	180
进样口温度/℃	180
热导池温度/℃	200
桥电流/mA	110
氢气压力/MPa	0.01
进样体积/μl	1
色谱柱类型	Porapak-Q

批次	加热功率/kW	原料乙醇质量分数	回流比	备注
1	1.75~2.50	0.429	1.0	调节加热功率使蒸汽率为5.6 L/h
2	1.70~2.50	0.492	1.0
3	1.65~2.50	0.589	1.0
4	1.70~2.50	0.652	1.0	调节加热功率使蒸汽率为6.5 L/h
5	1.73~2.50	0.490	1.5	调节加热功率使蒸汽率为5.6 L/h
6	1.70~2.50	0.590	1.5	调节加热功率使蒸汽率为5.6 L/h
7	1.75~2.25	0.650	1.5	调节加热功率使蒸汽率为6.5 L/h
8	1.75~2.50	0.493	2.0	调节加热功率使蒸汽率为5.6 L/h
9	1.70~2.50	0.592	2.0	调节加热功率使蒸汽率为5.6 L/h
10	1.78~2.25	0.654	2.0	调节加热功率使蒸汽率为6.5 L/h
11	1.77~2.50	0.486	2.5	调节加热功率使蒸汽率为5.6 L/h
12	1.72~2.50	0.614	2.5	调节加热功率使蒸汽率为5.6 L/h
13	1.75~2.25	0.685	2.5	调节加热功率使蒸汽率为6.5 L/h
14	2.0~2.25	0.674	2.5	调节加热功率使蒸汽率为8.5 L/h
15	1.65~2.50	0.683	2.5	调节加热功率使蒸汽率为5.6 L/h
16	1.80~2.25	0.677	3.0	调节加热功率使蒸汽率为6.5 L/h
17	1.75~2.50	0.726	3.5	调节加热功率使蒸汽率为6.5 L/h
18	1.60~2.50	0.483	3.3	恒定回流比生产操作，调节加热功率使蒸汽率为5.0 L/h
19	1.60~2.50	0.463	1.0 ~ 3.0	阶跃回流比生产操作，调节加热功率使蒸汽率为5.0 L/h

批次	加热功率/kW	原料乙醇质量分数	回流比	备注
1	1.75~2.50	0.429	1.0	调节加热功率使蒸汽率为5.6 L/h
2	1.70~2.50	0.492	1.0
3	1.65~2.50	0.589	1.0
4	1.70~2.50	0.652	1.0	调节加热功率使蒸汽率为6.5 L/h
5	1.73~2.50	0.490	1.5	调节加热功率使蒸汽率为5.6 L/h
6	1.70~2.50	0.590	1.5	调节加热功率使蒸汽率为5.6 L/h
7	1.75~2.25	0.650	1.5	调节加热功率使蒸汽率为6.5 L/h
8	1.75~2.50	0.493	2.0	调节加热功率使蒸汽率为5.6 L/h
9	1.70~2.50	0.592	2.0	调节加热功率使蒸汽率为5.6 L/h
10	1.78~2.25	0.654	2.0	调节加热功率使蒸汽率为6.5 L/h
11	1.77~2.50	0.486	2.5	调节加热功率使蒸汽率为5.6 L/h
12	1.72~2.50	0.614	2.5	调节加热功率使蒸汽率为5.6 L/h
13	1.75~2.25	0.685	2.5	调节加热功率使蒸汽率为6.5 L/h
14	2.0~2.25	0.674	2.5	调节加热功率使蒸汽率为8.5 L/h
15	1.65~2.50	0.683	2.5	调节加热功率使蒸汽率为5.6 L/h
16	1.80~2.25	0.677	3.0	调节加热功率使蒸汽率为6.5 L/h
17	1.75~2.50	0.726	3.5	调节加热功率使蒸汽率为6.5 L/h
18	1.60~2.50	0.483	3.3	恒定回流比生产操作，调节加热功率使蒸汽率为5.0 L/h
19	1.60~2.50	0.463	1.0 ~ 3.0	阶跃回流比生产操作，调节加热功率使蒸汽率为5.0 L/h

序号	特征		MIC	特征		MIC
序号	X	Y₁	MIC	X	Y₂	MIC
1	塔釜温度	馏出液乙醇浓度	0.3	塔釜温度	塔釜乙醇浓度	0.7
2	塔节一温度	馏出液乙醇浓度	0.3	塔节一温度	塔釜乙醇浓度	0.7
3	塔节三温度	馏出液乙醇浓度	0.5	塔节三温度	塔釜乙醇浓度	0.4
4	塔节五温度	馏出液乙醇浓度	0.7	塔节五温度	塔釜乙醇浓度	0.2
5	塔节六温度	馏出液乙醇浓度	0.5	塔节六温度	塔釜乙醇浓度	0.2
6	塔顶温度	馏出液乙醇浓度	0.4	塔顶温度	塔釜乙醇浓度	0.2
7	产品温度	馏出液乙醇浓度	0.4	产品温度	塔釜乙醇浓度	0.1
8	冷凝水流量	馏出液乙醇浓度	0.08	冷凝水流量	塔釜乙醇浓度	0.1
9	馏出液体积	馏出液乙醇浓度	0.2	馏出液体积	塔釜乙醇浓度	0.2
10	原料浓度	馏出液乙醇浓度	0.4	原料浓度	塔釜乙醇浓度	0.2
11	回流比	馏出液乙醇浓度	0.4	回流比	塔釜乙醇浓度	0.1
12	塔釜液位	馏出液乙醇浓度	0.2	塔釜液位	塔釜乙醇浓度	0.4
13	塔釜压力	馏出液乙醇浓度	0.4	塔釜压力	塔釜乙醇浓度	0.1
14	塔节二压力	馏出液乙醇浓度	0.4	塔节二压力	塔釜乙醇浓度	0.1
15	塔节四压力	馏出液乙醇浓度	0.5	塔节四压力	塔釜乙醇浓度	0.1
16	塔节七压力	馏出液乙醇浓度	0.4	塔节七压力	塔釜乙醇浓度	0.1