支持向量机与机理模型相结合的混合建模方法及其在聚酯生产过程的应用

以过程机理模型为基础,结合支持向量机,给出了一种适用于复杂非线性大迟滞过程的混合建模方法。该方法既保证了按照动力学规律描述过程特性,又充分利用现场运行和分析的数据,辩识模型结构参数。工程实际应用结果证明了这种模型的优良性能。     

聚丙烯腈质量指标软测量混合模型的应用

运用多种软测量技术，提出一种多类型混合软测量模型，并用于聚丙烯腈生产过程质量指标的推理估计，证明该模型性能优良。     

基于深度学习的聚合化学反应不确定热力学参数智能控制新法

聚合反应是个剧烈的放热反应，具有非线性、时变、大滞后的问题；聚合反应过程中会产生不确定的热力学参数问题，由于热力学参数问题难以解决，传统的方法比较繁琐和困难，聚合反应过程采用的模型预测的控制方法，将不确定的热力学参数上下进行取值，以获得更多的数据，运行出多组结果。将不确定的热力学参数作为输入，运行时间作为输出，得出多组实验后，再运用深度学习BP神经网络的方法给这些数据训练出来，得到一个模型，为了验证模型的准确性，把不确定的热力学参数代到目标函数中，得出的结果数据与实验的结果数据大致相符，进而模型的准确性得到验证。深度学习能够有效、简便地实现不确定参数的估计，从而实现整个反应过程控制的完成，这不仅能够提高效率，还能增加安全性。     

基于混合建模技术的聚氯乙烯粒径分布预测

针对聚氯乙烯粒径分布在线软测量问题,提出了一种基于机理分析和神经网络的混合建模方法,并将该建模方法应用于聚氯乙烯粒径分布建模研究中。混合模型由机理模型和误差补偿模型所组成。通过机理分析建立氯乙烯悬浮聚合过程的单体液滴群体平衡(Population Balance Equation,简称PBE)模型,由于聚氯乙烯成粒过程的复杂性和强非线性,单纯的机理模型预测与实际分析值相比仍存在一定偏差,因此利用人工神经网络建模方法建立了基于BP神经网络的单体液滴群体平衡模型修正模型,对单体液滴群体平衡模型的输出进行修正,由此建立起聚氯乙烯粒径分布混合模型。由于混合模型既能按照液滴分散与聚并机理对聚氯乙烯颗粒的成长过程进行描述,同时又充分利用了生产现场数据对模型误差进行修正,应用到聚氯乙烯生产过程的测试结果表明,与单纯机理模型相比,聚氯乙烯粒径分布混合模型具有更佳的预测效果。     

基于改进Kalman滤波算法的多模型融合建模方法

针对聚合物生产过程重要质量控制指标或状态变量的软测量问题,提出了一种基于改进Kalman滤波算法的多模型融合建模方法。将混合核函数主元分析(K₂PCA)与人工神经网络(ANN)相结合,建立一种基于K₂PCA-ANN的数据驱动模型;利用改进Kalman滤波算法实现K₂PCA-ANN模型与机理模型融合,构建一种并联结构的混合模型;协调二次滤波(线性滑动平滑)和方差更新对混合模型进行优化处理,使混合模型的估计性能尽可能地达到最优,使混合模型的预测稳定性得到有效改善。将该多模型融合建模方法应用于氯乙烯聚合过程聚合速率软测量中,应用研究结果表明:与单一的机理模型或K₂PCA-ANN数据驱动模型的预测性能相比,该建模方法建立的聚合速率模型具有更佳的预测性能。该建模方法的运用为进一步开展聚合物生产过程优化与控制等研究提供基础条件。     

自由基聚合物反应过程的超实时动态模拟

针对自由基聚合反应过程提出了一种简捷地建立其机理模型析新方法－链节分析法，解决了以往须同时求解无限多个微分方程才能模拟聚合过程的难。应用此方法，可以准确地描述了过程的动态特性和高聚物的平均分子量，结构信息，转化率等工业上关心的各种工艺指标，文中以低密度高压聚乙烯为例作了详细的说明。     

自由基聚合反应过程的超实时动态模拟

针对自由基聚合反应过程提出了一种简捷地建立其机理模型的新方法──链节分析法，解决了以往须同时求解无限多个微分方程才能模拟聚合过程的难题。应用此方法，可以准确地描述出过程的动态特性和高聚物的平均分子量、结构信息、转化率等工业上关心的各种工艺指标。文中以低密度高压聚乙烯为例作了详细的说明。     

生物反应过程的在线支持向量机建模优化

针对生物反应过程具有较强的非线性、时变性,建立准确的机理模型较为困难,并且复杂的机理模型也无法用于在线控制和优化。将在线支持向量机和机理模型结合,提出串并联在线自校正混合建模方法。通过对典型生化过程谷氨酸的生产过程分析,找到影响谷氨酸浓度的关键参数;从现场历史数据中选取样本,建立基于在线向量机的软测量模型。实验结果表明该模型对谷氨酸浓度预测效果较好。     

聚合工程研究动向

概况 聚合反应工程(简称聚合工程)系以工业规模的聚合过程为对象,以聚合动力学和聚合物系传递为基础,研究聚合反应器放大和聚合过程的开发(含仿真、优化及控制)。由此可掌握各类聚合反应过程的特性,建立操作参数和聚合物质量指标间的定量关系或模型化。     

基于核偏鲁棒M-回归的间歇反应过程混合模型辨识

提出了一种混合模型两步辨识策略,用以解决间歇反应过程的建模问题,并能够有效融合先验知识及过程数据信息。该策略将混合模型的同步辨识分解成为两个独立的步骤,首先确定混合模型的结构,并利用Tikhonov正则化方法实现间歇反应过程反应速率的精确估计;接下来采用核偏鲁棒M-回归(kernel partial robust M-regression,KPRM)算法建立过程变量与反应速率间的经验模型,从而有效抑制过程数据中离群点的影响。利用半间歇过程仿真实验对所提出的策略进行验证,获得了相比于传统方法更高的估计及预测精度。     

基于动态过程划分的熔融指数软测量建模

牌号切换过程是聚合反应过程节能降耗的关键环节,已有的模型往往因为质量样本较少而弱化甚至避开这一过程。因此,在之前专门针对牌号切换过程提出的三阶段分解方法的基础上,进一步面向整个生产过程,针对聚丙烯反应过程存在的同一牌号的稳定生产过程以及不同牌号间的切换过程具有不同动态过程的特性,按照不同的生产模式和生产牌号划分不同动态过程的样本,采用多模型的方法在各自的样本集上建立子模型,有针对性地把握相应的动态变化规律。为了实现多个子模型之间的切换,进一步基于反应条件和反应结果估计值构建了综合判断模型。最后,通过实际数据验证,三阶段多模型相对Kim多模型、单一模型来说,具有更好的预测结果。     

基于混合建模的水泥生料分解过程动态特性研究

为掌握水泥分解炉运行过程的动态特性,采用机理建模与神经网络相结合的方法构建了水泥分解炉一维特性模型,并结合工业数据对该方法的可行性进行验证。结果表明,模型能够准确地计算炉内温度、气体浓度等参数,具有良好的泛化性能。基于所提出的模型,研究了炉内各状态参数的稳态分布特性。此外,对喷煤量、生料下料量、喷氨量以及高温风机转速等操作变量进行阶跃实验,分析上述操作变量改变时分解炉出口温度及出口NO _x 含量的动态响应情况。研究所得相关动态特性规律可以为控制系统的分析、设计和优化提供参考与依据。     

基于混合模型的气流床气化炉建模

为了提高在煤质改变及工艺参数波动条件下气流床气化炉出口结果的预测精度,分别采用机理模型、广义回归神经网络（GRNN）模型以及混合模型对气化炉进行建模,其中混合模型由GRNN模型和机理模型构建,结合两种不同的煤样对三种模型的预测结果进行分析。结果表明：三种模型均可以较好地对气化过程进行模拟;其中在煤种固定的情况下混合模型关于气化温度和CO、CO₂及H₂含量的预测误差为0.18%和0.25%、1.72%及0.43%,与机理模型和GRNN模型相比误差更小;在煤种改变的情况下混合模型关于出口气体结果的预测最接近实际生产数据,误差为0.81%和0.11%、2.53%及0.42%。证明混合模型在煤种改变及工艺参数波动条件下可以有效地对气化过程进行模拟,在很大程度上提高了机理模型和GRNN模型的预测精度。     

A model-based characterization of the long-term asymptotic behavior of nonlinear discrete-time processes using invariance functional equations

The present research work proposes a new approach to the problem of quantitatively characterizing the long-term dynamic behavior of nonlinear discrete-time processes. It is assumed that in order to analyze the process dynamic behavior and digitally simulate it for performance monitoring purposes, the discrete-time dynamic process model considered can be obtained: (i) either through the employment of efficient and accurate discretization methods for the original continuous-time process which is mathematically described by a system of nonlinear ordinary (ODEs) or partial differential equations (PDEs) or (ii) through direct identification methods. In particular, nonlinear processes are considered whose dynamics can be viewed as driven: (i) either by an external time-varying “forcing” input/disturbance term, (ii) by a set of time-varying process parameters or (iii) by the autonomous dynamics of an upstream process. The formulation of the problem of interest can be naturally realized through a system of nonlinear functional equations (NFEs), for which a rather general set of conditions for the existence and uniqueness of a solution is derived. The solution to the aforementioned system of NFEs is then proven to represent a locally analytic invariant manifold of the nonlinear discrete-time process under consideration. The local analyticity property of the invariant manifold map enables the development of a series solution method for the above system of NFEs, which can be easily implemented with the aid of a symbolic software package such as MAPLE. Under a certain set of conditions, it is shown that the invariant manifold computed attracts all system trajectories, and therefore, the asymptotic process response and long-term dynamic behavior are determined through the restriction of the discrete-time process dynamics on the invariant manifold.     

Inferential sensors for estimation of polymer quality parameters: Industrial application of a PLS-based soft sensor for a LDPE plant

Low-density polyethylene (LDPE) and ethylene vinyl acetate (EVA) copolymers are produced in free radical polymerization using reactors at extremely high pressure. The reactors require constant monitoring and control in order to minimize undesirable process excursions and meet stringent product specifications. In industrial settings, polymer quality is mainly specified in terms of melt flow index (MI) and density. These properties are difficult to measure and usually unavailable in real time, which leads to major difficulty in controlling product quality in polymerization processes. Researchers have attempted first principles modeling of polymerization processes to estimate end use properties. However, development of detailed first principles model for free radical polymerization is not a trivial task. The difficulties involved are the large number of complex and simultaneous reactions and the need to estimate a large number of kinetic parameters. To overcome these difficulties, some researchers considered empirical neural network models as an alternative. However, neural network models provide no physical insight about the underlying process. We consider data-based multivariate regression methods as alternative solution to the problem. In this paper, some recent developments in modeling polymer quality parameters are reviewed, with emphasis given to the free radical polymerization process. We present an application of PLS to build a soft-sensor to predict melt flow index using routinely measured process variables. Issues of data acquisition and preprocessing for real industrial data are discussed. The study was conducted using data collected form an industrial autoclave reactor, which produces LDPE and EVA copolymer using free radical polymerization. The results indicated that melt index (MI) can be successfully predicted using this relatively straightforward statistical tool.     

Optimal hybrid modeling approach for polymerization reactors using parameter estimation techniques

The dynamics of polymerization catalytic reactors have been investigated by many researchers during the past five decades; however, the emphasis of these studies was directed towards correlating process model parameters using empirical investigation based on small scale experimental setup and not on real process conditions. The resulting correlations are of limited practical use for industrial scale operations. A statistical study for the relative correlation of each of the effective process parameters revealed the best combination of parameters that could be used for optimizing the process model performance. Parameter estimation techniques are then utilized to find the values of these parameters that minimize a predefined objective function. Published real industrial scale data for the process was used as a basis for validating the process model. To generalize the model, an artificial neural network approach is used to capture the functional relationship of the selected parameters with the process operating conditions. The developed ANN-based correlation was used in a conventional fluidized catalytic bed reactor (FCR) model and simulated under industrial operating conditions. The new hybrid model predictions of the melt-flow index and the emulsion temperature were compared to industrial measurements as well as published models. The predictive quality of the hybrid model was superior to other models. The suggested parameter estimation and modeling approach can be used for process analysis and possible control system design and optimization investigations.     

A framework of hybrid model development with identification of plant-model mismatch

Hybrid modeling has attracted increasing attention in order to take advantage of the additional data to improve process understanding. Current practice often adopts mechanistic models to predict process behaviors. These mechanistic models are based on physical understandings and experimental studies, but they sometimes lead to plant-model mismatch (PMM) as they may be inaccurate to fully describe real processes. Black-box models can serve as an alternative, but they often suffer from poor generalization and interpretability. To combine the two techniques, hybrid models are developed to make use of process data while maintaining a degree of physical understanding. In this work, we implement a framework of identification of PMM using partial correlation coefficient and mutual information, followed by introducing and comparing serial, parallel, and combined structures of hybrid models. The framework is applied and tested with a simulated reactor model and two pharmaceutical unit operation case studies.     

融合过程先验知识的递归神经网络模型及其应用

大部分化工过程具有非线性特性,一般的线性建模方法难以有效应用。针对非线性化工过程动态建模,提出了一种基于过程先验知识的递归神经网络模型,充分发掘化工过程隐含的先验知识,并将这些先验知识以非线性约束的形式嵌入NARMAX结构的前馈神经网络中,同时基于增广拉格朗日乘子法约束处理机制,用PSO-IPOPT混合优化算法对过程先验知识递归神经网络权值进行优化。该过程先验知识递归神经网络模型对非线性化工过程动态建模,不仅有良好的建模精度和预测外推能力,而且能避免零增益的出现和增益反转,确保网络模型在实际应用中的安全性。文中以环管式丙烯聚合反应过程实际工业数据验证了所提网络模型的有效性。     

A hybrid science-guided machine learning approach for modeling chemical processes: A review

This study presents a broad perspective of hybrid process modeling combining the scientific knowledge and data analytics in bioprocessing and chemical engineering with a science-guided machine learning (SGML) approach. We divide the approach into two major categories: ML complements science, and science complements ML. We review the literature relating to the hybrid SGML approach, and propose a systematic classification of hybrid SGML models. For applying ML to improve science-based models, we present expositions of direct serial and parallel hybrid modeling and their combinations, inverse modeling, reduced-order modeling, quantifying uncertainty in the process and even discovering governing equations of the process model. For applying scientific principles to improve ML models, we discuss the science-guided design, learning and refinement. For each subcategory, we identify its requirements, strengths, and limitations, together with their published and potential applications. We also present several examples to illustrate different hybrid SGML methodologies for modeling chemical processes.