1.School of Mechanical Engineering,North University of China,Taiyuan 030051,China
2.School of Information and Control Engineering,China University of Mining and Technology,Xuzhou 221116,China
3.School of Software,North University of China,Taiyuan 030051,China
4.Changshan County Shuangming Bearing Co. ,Ltd. ,Quzhou 324299,China
5.School of Mechanical Engineering,China University of Mining and Technology,Xuzhou 221116,China
6.Key Laboratory of Computing Power Network and Information Security,Ministry of Education,Jinan 250014,China
Show less
文章历史+
Received
Accepted
Published
2026-04-13
2026-06-16
2026-10-31
Issue Date
2026-10-28
PDF (2299K)
摘要
针对中药生产因药材批次差异、 蒸汽压力波动等因素导致的蒸汽调节阀控制滞后性问题, 以及传统控制方法难以兼顾产品质量与能耗的痛点, 提出了一种基于主动学习的带外部输入非线性自回归(Nonlinear Autoregressive with Exogenous Inputs Based on Active Learning, NARX-AL)模型预测控制系统。该系统以关键质量属性(Critical Quality Attribute, CQA)的实时数据为基础, 利用主动学习策略对有限批次数据进行高效重构, 构建NARX动态预测模型, 并设计模型预测控制器, 在满足开度、 变化率及CQA区间约束下优化蒸汽调节阀开度。以心可舒片醇提工序为对象, 在ZQ-100N中药提取浓缩设备上开展验证实验。结果表明, 基于NARX-AL的模型预测控制系统能够将CQA含量稳定在预设区间内, 优于固定参数控制和PID (Proportion Integration Differentiation)控制, 蒸汽能耗相比固定参数控制降低10%以上, 且全过程符合监管要求。该方法为中药生产过程的精细化、 智能化控制及能耗优化提供了可行的技术路径。
Abstract
A predictive control system based on a nonlinear autoregressive with exogenous input and active learning (NARX-AL) model is proposed to address the issues of control lag in steam regulating valves caused by factors such as batch differences in Chinese herbal materials and steam pressure fluctuations in traditional Chinese medicine (TCM) production, as well as the pain points of traditional control methods that struggle to balance product quality and energy consumption. Based on real-time data of critical quality attributes (CQA), active learning strategies are employed to efficiently reconstruct limited batch data, a NARX dynamic predictive model is constructed, and a model predictive controller (MPC) is designed to optimize the regulating valve opening under the constraints of opening, rate of change, and CQA range. Taking the ethanol extraction process of Xinkeshu Tablets as the object, verification experiments are conducted on the ZQ-100N TCM extraction and concentration equipment. The results show that the NARX-AL-based model predictive control system is able to stabilize the CQA content within a preset range, outperforming fixed parameter control and PID control. Steam energy consumption is reduced by more than 10% compared to fixed parameter control, and the entire process is shown to meet regulatory requirements. The method proposed in this paper provides a feasible technical path for refined and intelligent control and energy consumption optimization in TCM production processes.
模型预测控制(Model Predictive Control, MPC)作为一种先进的优化控制策略, 在化工、 制药等过程工业中得到了广泛应用[13]。MPC通过建立被控对象的预测模型, 对系统未来一段时间内的输出进行滚动预测, 并通过在线求解带约束的有限时域优化问题获得最优控制序列, 能够显式处理输入输出约束、 多变量耦合以及纯滞后等问题, 适合具有约束条件的工业过程控制问题[14]。然而, MPC的控制性能在很大程度上依赖于预测模型的精度, 当被控对象具有强非线性和工况变化特性时, 其控制效果会明显下降[15]。此外, 现有MPC框架缺乏对模型参数和控制器参数的在线自适应能力, 且与《药品生产质量管理规范》(Good Manufacturing Practice of Medical Products, GMP)对变更可追溯、 干预可解释的要求存在冲突。
针对上述问题, 本文提出了基于主动学习的带外部输入非线性自回归(Nonlinear Auto Regressive with Exogenous Inputs Based on Active Learning, NARX-AL)模型预测控制系统, 该系统以已获取的关键质量属性(Critical Quality Attribute, CQA)实时数据为基础, 引入主动学习思想对有限批次数据进行重构, 建立基于NARX的动态预测模型, 在满足监管约束的情况下, 实现中药生产过程质量的前馈优化控制。
HuangLuqi. Consideration on the key directions for fundamental research in science and technology innovation of traditional Chinese medicine[J]. Bulletin of National Natural Science Foundation of China, 2024, 38(3): 383-386. (in Chinese)
GuanYongmei, LiuLeling, YangMing, et al. Key issues and application progress in scaling up traditional Chinese medicine extraction processes based on process analytical technology[J]. Chinese Traditional and Herbal Drugs, 2026, 57(2): 767-777. (in Chinese)
ChengHaibo, ZhangLei, FuYong, et al. 2023 major scientific issues, engineering challenges and industrial technology problems[J]. Journal of Traditional Chinese Medicine, 2023, 64(14): 1405-1421. (in Chinese)
[7]
张玉雪.中药提取设备温度和压力在线控制研究[D].济南: 山东大学, 2022: 1-63.
[8]
ZhangYuxue.Research on temperature and pressure control of tradition Chinese medicine extraction equipment[D].Jinan: Shandong University, 2022: 1-63. (in Chinese)
ChenWeiji. Application of self-adaptive PID control in energy-saving system of cold and hot station[J]. Journal of Test and Measurement Technology, 2022, 36(1): 60-65. (in Chinese)
[11]
GamsM, HorvatM, OžekM, et al. Integrating artificial and human intelligence into tablet production process[J]. AAPS PharmSciTech, 2014, 15(6): 1447-1453.
[12]
YanB, LiY, GuoZ, et al. Quality by design for herbal drugs: a feedforward control strategy and an approach to define the acceptable ranges of critical quality attributes[J]. Phytochemical Analysis, 2014, 25(1): 59-65.
[13]
Garcı́A-MuñozS, MercadoJ. Optimal selection of raw materials for pharmaceutical drug product design and manufacture using mixed integer nonlinear programming and multivariate latent variable regression models[J]. Industrial & Engineering Chemistry Research, 2013, 52(17): 5934-5942.
[14]
SinghR, SahayA, KarryK M, et al. Implementation of an advanced hybrid MPC-PID control system using PAT tools into a direct compaction continuous pharmaceutical tablet manufacturing pilot plant[J]. International Journal of Pharmaceutics, 2014, 473(1/2): 38-54.
[15]
PrikerznikM, SrcicS. Multivariate analysis for optimization and validation of the industrial tablet-manufacturing process[J]. Drug Development and Industrial Pharmacy, 2021, 47(1): 61-71.
[16]
RadcliffeA J, ReklaitisG V. Process monitoring under uncertainty: an opportunity for Bayesian multilevel modelling[M]//31st European Symposium on Computer Aided Process Engineering. Amsterdam: Elsevier, 2021: 1377-1382.
[17]
SunL, XiJ, XiaQ, et al. Optimal control strategy of multi-stage pharmaceutical[J]. IFAC-PapersOnLine, 2019, 52(10): 370-375.
HuangHaisong, XieQingsheng, LiShaobo. Predictive control-based control strategy for multi-objective optimization of extraction section of Chinese traditional medicine[J]. Journal of Lanzhou University of Technology, 2013, 39(3): 75-79. (in Chinese)
[20]
Villa-TamayoM F, CaicedoM A, RivadeneiraP S. Offset-free MPC strategy for nonzero regulation of linear impulsive systems[J]. ISA Transactions, 2020, 101: 91-101.
ZhangAng, ZhouChao, LiuLin. Research on height tracking controller of corn header based on model predictive control[J]. Modern Agricultural Science and Technology, 2026(5): 135-138. (in Chinese)
[23]
ChenZ, ShenY, ChenB, et al. Vasline: realize online detection and augmented NIR using deep learning[J]. Engineering Applications of Artificial Intelligence, 2023, 126: 106684.