基于模型性能评估的递推PLS建模及应用

为了降低递推部分最小二乘(RPLS)建模方法的模型校正频率,开发了一种基于模型性能评估的RPLS(MPA-RPLS)模型.首先,根据过程的初始特性,自动生成模型的置信限,以均方根误差(RMSEP)为性能指标,评估模型性能;依据模型性能的评估结果,选择性地启动模型校正和置信限校正.然后,引入滑动平均滤波器消除过程变量中的噪声,探讨噪声对模型性能的影响程度.最后,将MPA-RPLS模型应用于一个化学反应过程--C₈芳烃临氢异构化过程,基于大量工业数据,进行仿真验证.仿真结果表明:本文开发的模型仅以微小的精度损失换取了模型计算效率的大幅提高(即模型校正频率大幅下降);滑动平均滤波器可有效地处理变量的噪声,改善模型的预测精度.     

移动窗递推PLS软测量建模及其工业应用

讨论了基于递推部分最小二乘(RPLS)算法的软测量建模方法及其应用.针对过程的时变特性,采用移动窗RPLS算法,在线修正样本的均值和方差,实时更新模型参数,兼顾了建模样本的饱和性、样本信息的完整性.然后将软测量模型应用于工业异构化装置,在线估计对二甲苯(PX)的含量.针对大量工业数据,进行仿真计算,得到模型的最大相对误差、相对均方误差和跟踪性能指标分别为2.68%、0.17%和0.9569,说明该软测量模型具有良好的预测能力和跟踪性能.接着讨论了建模样本长度对模型性能的影响,指出其最佳的样本长度为20～50.     

Adaptive Nonlinear Model Predictive Control Using an On-line Support Vector Regression Updating Strategy

The performance of data-drivenmodels relies heavily on the amount and quality of training samples, so itmight deteriorate significantly in the regions where samples are scarce. The objective of this paper is to develop an online SVR model updating strategy to track the change in the process characteristics efficiently with affordable computational burden. This is achieved by adding a new sample that violates the Karush-Kuhn-Tucker conditions of the existing SVR model and by deleting the old sample that has the maximum distance with respect to the newly added sample in feature space. The benefits offered by such an updating strategy are exploited to develop an adaptive model-based control scheme, where model updating and control task perform alternately. The effectiveness of the adaptive controller is demonstrated by simulation study on a continuous stirred tank reactor. The results reveal that the adaptive MPC scheme outperforms its non-adaptive counterpart for largemagnitude set point changes and variations in process parameters.     

基于递推部分最小二乘自适应质量监控策略及其在橡胶混炼过程中的应用

提出一种新的基于递推部分最小二乘（RPLS）算法的自适应在线质量监控策略。利用隐变量选择算法,根据实时采集的现场数据,在不增加计算和存储容量的基础上递推更新RPLS过程监测模型,进而更新Qα控制限,从而使RPLS自适应质量监控系统具有强时变跟踪特性,能够有效克服传统监测算法Qα无法反映系统时变性的缺点,大大降低了监控系统的误报率和漏报率,提高监控系统性能。并根据橡胶混炼过程特点,将此方法运用于该时变间歇过程质量监控中,取得了满意效果。     

A novel Q-based online model updating strategy and its application in statistical process control for rubber mixing

To overcome the large time-delay in measuring the hardness of mixed rubber, rheological parameters were used to predict the hardness. A novel Q-based model updating strategy was proposed as a universal platform to track time-varying properties. Using a few selected support samples to update the model, the strategy could dramat-ical y save the storage cost and overcome the adverse influence of low signal-to-noise ratio samples. Moreover, it could be applied to any statistical process monitoring system without drastic changes to them, which is practical for industrial practices. As examples, the Q-based strategy was integrated with three popular algorithms (partial least squares (PLS), recursive PLS (RPLS), and kernel PLS (KPLS)) to form novel regression ones, QPLS, QRPLS and QKPLS, respectively. The applications for predicting mixed rubber hardness on a large-scale tire plant in east China prove the theoretical considerations.     

Staged Versions of Liquid Membrane Extraction Processes

The application of the liquid membrane technique to the system consisting of serially connected mixer-settler extractors is suggested and evaluated. A comparative theoretical analysis of various flow schemes of contacting liquids in series of mixer-settler extractors has been performed. It is shown that the scheme of a staged version of supported liquid membrane provides the highest extraction efficiency. One laboratory plant consisting of eight mixer-settler was examined with the standard extraction system water-acetone-toluene using two schemes: with internal (the extractant recycles between each pair of extraction and stripping stages) and external (the extractant first passes through all the stages of extraction, then through all stripping stages) circulation of the extractant. The effectiveness of the first variant (supported liquid membrane scheme) was considerably higher than that of the second scheme. Good agreement between theory and experiment was established.     

Comparative study of estimation methods of NO<Subscript><Emphasis Type="Italic">x</Emphasis></Subscript> emission with selection of input parameters for a coal-fired boiler

This study focuses on estimation of NO_x emission and selection of input parameters for a coal-fired boiler in a 500 MW power generation plant. Careful selection of input parameters is required not only to improve accuracy of the estimation, but also to reduce the model dimensionality. The initial operating input parameters are determined based on operation heuristics and accumulated operation knowledge; the essential input parameters are selected by sensitivity analysis where the performance of the estimation model is assessed as one or some input parameters are successively eliminated from the computation while all other input parameters are retained. From the sequential input selection process, less than ten input parameters survived out of 36 initial input parameters. Auto-regressive moving average (ARMA) model, artificial neural networks (ANN), partial least-squares (PLS) model, and least-squares support vector machine (LSSVM) algorithm were proposed to express the relationship between the operating input parameters and the content of NO_x emission. Historical real-time data obtained from a 500 MW power plant coal-fired boiler were used to test the proposed models. It was found that principal components analysis (PCA) enhances the estimation performance of each model. Among the four proposed estimation models, the LSSVM model coupled with PCA scheme showed the minimum root-mean square error (RMSE) and the best R-square value.     

Assessment and comparison of distributed model predictive control schemes: Application to a natural gas refrigeration plant

A number of decentralized and distributed control schemes based on model predictive control (MPC) have been introduced in the last years. They have been proposed as viable solutions to the computational, transmission and robustness issues arising in the centralized context in case of large-scale and/or distributed plants. Such MPC-based control schemes are very heterogeneous, based on different model structures and realizations, with different features and infrastructural/memory/computational requirements.In this paper, we test and compare, with a realistic case study, a robust non-cooperative scheme and a cooperative iterative one. The main scope is to analyze and unravel, in a fair comparison scenario, these methods from different viewpoints, spanning from the model realization issues to the communication and computational requirements, to the control performances. The benchmark case study consists of an existing natural gas refrigeration plant. Realistic simulations and validation tests are obtained through in the DynSim industrial process simulation environment.     

拟建环氧丙烷装置多方案技术经济比较和风险分析

对拟建环氧丙烷装置的多个工艺方案,从技术经济的角度和各方案存在的主要风险进行了分析和阐述,并推荐适用可行的方案,帮助业主在工艺技术路线选择时进行正确的决策。     

DYNAMIC AND HYBRID NEURAL MODEL OF THERMAL DRYING IN A FLUTDIZED BED

A preliminary study aimed at comparing Classical Dynamic Neural Modelling (CDNM) and Hybrid Neural Modelling (HNM) to describe thermal dewatering process in a fluidized bed is presented Two schemes of HN modelling were developed to find the most efficient way of combining a classical mathematical model of the process and Artificial Neural Network (ANN). CDN model was developed using “moving window” technique. In the first scheme of HNM a feed-forward ANN was trained to predict evaporation rate and heat flux in the drying process. In the second scheme of the HN model, ANN was used to determine heat transfer coefficient only. Excellent prediction of drying process by HNM is proved.     

Effect of Mixing Schemes on Physico-Mechanical,Aging Properties,and Permeability of Aldehydes through Blends of Natural Rubber and Low Molecular Weight Natural Rubber

This article reports on the physico-mechanical properties, chemical resistance, aging properties, sorption, diffusion, and permeability of aldehydes (acetaldehyde and formaldehyde) through vulcanizates from blends of natural rubber (NR) and low molecular weight natural rubber (LMWNR) compounded by three different mixing schemes. The compounding ingredients were mixed with the two mentioned rubbers using three different mixing schemes by adopting the semi-efficient sulphur vulcanization compounding formulation. In scheme 1, the natural rubber and LMWNR were first mixed before adding the compounding ingredients. In scheme 2, the compounding ingredients were first mixed with the NR before adding the LWMNR and in scheme 3, the compounding ingredients were first mixed with the LMWNR before adding the NR. The physico-mechanical results of the vulcanizates showed that changes in the mixing schemes significantly influence the tensile properties of the vulcanizates. The tensile strengths of the vulcanizates prepared with mixing scheme 2 were 3.5 MPa lower than vulcanizates from scheme 1, whereas scheme 3 was lower than scheme 1 by 7.8 MPa. The aging result of the vulcanizates from all the mixing schemes were found to be similar. The activation energy and free energy change were highest with scheme 1 whereas the extent of acetaldehyde and formaldehyde penetrations were lowest with scheme 1, signifying a well crosslinked and aldehyde-resistant vulcanizate.     

DYNAMIC AND HYBRID NEURAL MODEL OF THERMAL DRYING IN A FLUTDIZED BED

ABSTRACT

A preliminary study aimed at comparing Classical Dynamic Neural Modelling (CDNM) and Hybrid Neural Modelling (HNM) to describe thermal dewatering process in a fluidized bed is presented Two schemes of HN modelling were developed to find the most efficient way of combining a classical mathematical model of the process and Artificial Neural Network (ANN). CDN model was developed using “moving window” technique. In the first scheme of HNM a feed-forward ANN was trained to predict evaporation rate and heat flux in the drying process. In the second scheme of the HN model, ANN was used to determine heat transfer coefficient only. Excellent prediction of drying process by HNM is proved.     

Adaptive monitoring statistics with state space model updating based on canonical variate analysis

A new multivariate statistical model updating by using a recursive state space model updating based on CVA is proposed. The CVA-based monitoring techniques have been researched to detect and isolate process abnormalities in dynamic processes. Two monitoring indices are defined for fault detection, and a state space model updating procedure is developed by using mean, covariance, and correlation updating based on forgetting factor as well as the recursive Cholesky factor updating. To adjust forgetting factors according to variation of process state, the forgetting factor updating criteria are introduced. The proposed method is applied to benchmark models of a continuous stirred tank reactor with a first order reaction and the Tennessee Eastman process (TEP) under transient and time-varying operating conditions. Through the simulation results, we expect that the proposed approach can be applied to time-varying and dynamic processes under transient state.     

EXTENDED KALMAN FILTER CONTROLLER: FIRST PRINCIPLES MODELS TO NEURAL NETWORKS

Extended Kalman filters (EKF) have been widely employed for state and parameter estimation in chemical engineering systems. Gao et al. [Gao, F., Wang, F. and Li, M. (1999). Ind. Eng. Chem. Res., 38, 2345-2349] have proposed the use of EKF for control computation using a neural network representation of the system in a discrete-time framework. In the present study, an EKF controller is proposed in a continuous time framework with models incorporating different levels of process knowledge. The problem of process-model mismatch is handled by incorporating EKF-based state and/or parameter estimation along with control computation. A batch reactor temperature control problem for a highly exothermic reaction between maleic anhydride and hexanol to form hexyl monoester of maleic acid is considered as a test bed to evaluate the performance of the proposed control schemes. Three different models are considered, namely the first principles model, a reduced-order process model, and an artificial neural network (ANN) model for formulation of the control schemes. The performance of the proposed control scheme using first principles model is compared to that of generic model control, and a similar performance is achieved. The present study illustrates the usefulness of the proposed control schemes and can be easily extended to general chemical engineering systems.     

煤气化技术方案的选择对煤制天然气项目中甲烷化装置的影响

介绍了煤制天然气项目的工艺流程及其甲烷化装置的反应原理;对比了分别采用Shell、KBR或Lugri煤气化技术方案时甲烷化装置的处理气量、循环量以及副产过热蒸汽产量。结果表明,在产品规模相同的情况下：①从投资角度分析,Shell方案甲烷化装置处理气量最大,但投资最高,KBR方案居中,Lugri方案最佳;②从运行成本角度分析,Shell方案副产蒸汽量最高,运行成本最低,其次是KBR方案,再次是Lugri方案。     

On the application of liquid-membrane principle in a system of mixing-settling extractors

The paper is devoted to the possibilities of applying the liquid-membrane technique in extraction processes in a cascade of mixing-settling apparatuses. The process is performed in the cascade of N mass-exchange stages, each of which contains two mixing-settling extractors, one of which is used in the extraction of a component from the starting water solution, while the other is used in the its reextraction by the second (pure) water solution. Calculation dependences between the input and output concentrations in flows are represented. Experimental investigations were carried out on a laboratory plant consisting of eight glass cylindrical mixing-settling apparatuses with a standard extraction system water—acetone-toluene at two schemes of the stage assemblage: with the internal (between the stages of extraction and reextraction) and external (extractant flow passes first through all extraction stages and next reextraction those) extractant circulation. The efficiency of the first variant (liquid membrane scheme) was found to be two times higher than that of the second variant.     

Modelling of adsorption of nickel (II) by blend hydrogels (cellulose nanocrystals and corn starch) from aqueous solution using adaptive neuro-fuzzy inference systems (ANFIS) and artificial neural networks (ANN)

The current study looks at the effectiveness of the removal of nickel (II) from aqueous solution using an adsorption method in a laboratory-size reactor. An artificial neural network (ANN) and an adaptive neuro-fuzzy inference system (ANFIS) were used in this study to predict blend hydrogels adsorption potential in the removal of nickel (II) from aqueous solution. Four operational variables, including initial Ni (II) concentration (mg/L), pH, contact duration (min), and adsorbent dose (mg/L), were used as an input with removal percentage (%) as the only output; they were studied to assess their impact on the adsorption study in the ANFIS model. In contrast, 70% of the data was used for training, while 15% of the data was used in testing and validation to build the ANN model. Feedforward propagation with the Levenberg–Marquardt algorithm was employed to train the network. The use of ANN and ANFIS models for experiments was used to optimize, construct, and develop prediction models for Ni (II) adsorption using blend hydrogels. The adsorption isotherm and kinetic models were also used to describe the process. The results show that ANN and ANFIS models are promising prediction approaches that can be applied to successfully predict metal ions adsorption. According to this finding, the root mean square errors (RMSE), absolute average relative errors (AARE), average relative errors (ARE), mean squared deviation (MSE), and R² for Ni (II) in the training dataset were 0.061, 0.078, 0.017, 0.019, and 0.986, respectively, for ANN. In the ANFIS model, the RMSE, AARE, ARE, MSE, and R² were 0.0129, 0.0119, 0.028, 0.030, and 0.995, respectively. The adsorption process was spontaneous and well explained by the Langmuir model, and chemisorption was the primary control. The morphology, functional groups, thermal characteristics, and crystallinity of blend hydrogels were all assessed.     

基于PLS-LSSVM的谷氨酸发酵产物浓度预测建模

针对谷氨酸发酵过程关键生化参数难以在线检测给发酵优化控制带来困难问题,基于谷氨酸5 L发酵罐发酵过程,建立基于偏最小二乘（PLS）和最小二乘向量机（LSSVM）相结合的谷氨酸浓度预测模型;利用PLS对输入变量进行特征提取降低维数和消除相关性,以简化模型和提高模型精度。为确定谷氨酸发酵最佳预测模型,简化后的预测模型与发酵动力学模型进行比较;实验结果表明,简化后的耦合模拟退火（coupled simulated annealing,CSA）对参数进行优化的LSSVM模型具有最好预测性能,相对PLS预测模型和发酵动力学模型具有明显优势,均方根误差分别为1.597、8.49和2.934,可以为谷氨酸发酵过程操作及时调整及优化控制提供有效指导。     

Nonlinear state estimation and control of a batch reactive distillation

This paper deals with the advanced adaptive control of a batch reactive distillation (RD) column for the production of ethyl acetate. The nonlinear adaptive control law consists of the generic model controller (GMC) and an adaptive state estimator (ASE). In the first part of the present work, the design approach of the ASE scheme in two different forms, namely ASE1 and ASE2, has been addressed for a batch reactive rectifier. The predictor model of both the ASE estimators includes only a component mole balance equation around the condenser-reflux drum system and an extra state equation having no dynamics, and therefore, there is a large process/predictor mismatch. In presence of this structural discrepancy, the adaptive estimation schemes compute the imprecisely known parameters quite accurately based on the measured distillate composition under initialization error, disturbance and uncertainty. In the subsequent part, the adaptive GMC–ASE1 control structure has been formulated for the sample reactive column. This nonlinear control strategy shows comparatively better closed-loop performance than the gain-scheduled proportional integral (GSPI) controller due to the exponential error convergence capability of the estimation scheme and the high-quality control of the GMC law.     

Numerical solution of population balance equations for nucleation, growth and aggregation processes

This article focuses on the derivation of numerical schemes for solving population balance models (PBMs) with simultaneous nucleation, growth and aggregation processes. Two numerical methods are proposed for this purpose. The first method combines a method of characteristics (MOC) for growth process with a finite volume scheme (FVS) for aggregation process. For handling nucleation terms, a cell of nuclei size is added at a given time level. The second method purely uses a semi-discrete finite volume scheme for nucleation, growth and aggregation of particles. Note that both schemes use the same finite volume scheme for aggregation process. On one hand, the method of characteristics offers a technique which is in general a powerful tool for solving linear growth processes, has the capability to overcome numerical diffusion and dispersion, is computationally efficient, as well as give highly resolved solutions. On the other hand, the finite volume schemes which were derived for a general system in divergence form, are applicable to any grid to control resolution, and are also computationally not expensive. In the first method a combination of finite volume scheme and the method of characteristics gives a highly accurate and efficient scheme for simultaneous nucleation, growth and aggregation processes. The second method demonstrates the applicability, generality, robustness and efficiency of high-resolution schemes. The proposed techniques are tested for pure growth, simultaneous growth and aggregation, nucleation and growth, as well as simultaneous nucleation, growth and aggregation processes. The numerical results of both schemes are compared with each other and are also validated against available analytical solutions. The numerical results of the schemes are in good agreement with the analytical solutions.