Control of an Industrial Distillation Column Using a Hybrid Model with Adaptation of the Range of Validity and an ANN-based Soft Sensor

Advanced control schemes such as model predictive control can be used to minimize the use of resources while guaranteeing the specified product quality. In this paper, we consider an industrial mother liquor distillation column varying flow rate and composition of the feed. There are specifications of the composition for all product streams. To address this challenging control problem, we employ a nonlinear model-predictive controller using a hybrid model, which consists of a simple phenomenological model augmented by a data-based component to compensate the plant-model mismatch. The trustworthiness of the data-based model is addressed using a domain of validity of the data-based model, which is estimated using a one-class support vector machine. During operation, it may turn out that the model is also reliable in a wider range, therefore, data of recently visited operating points is recorded and the domain of validity is extended if the model is sufficiently accurate. To improve the performance of the controller, an artificial neural network model is used to estimate the product composition from available measurements.     

应用经遗传算法优化的BP神经网络预测催化裂化装置焦炭产率

焦炭是催化裂化装置的主要副产物,准确预测催化裂化焦炭产率对提高装置的操作平稳度和经济效益具有重要意义。人工神经网络(ANN)具有强大的自学习和自适应能力,在非线性预测方面具有明显的优势。本研究将遗传算法(GA)与BP神经网络相结合,基于某炼厂催化裂化装置的生产数据,分别从原料、催化剂和操作条件3个方面选取28个关键影响参数建立了催化裂化焦炭产率预测模型,分别将BP神经网络和经遗传算法优化的BP神经网络(GA-BP)的预测结果与工业数据进行对比。结果表明,经遗传算法优化的预测模型无论在预测结果的准确性还是稳定性方面效果更好。最后,本研究还通过考察原料残炭、反应温度等单一关键参数对焦炭产率的影响,进一步证明了经遗传算法优化的BP神经网络预测模型的准确性。     

基于神经网络的软测量技术在精馏塔上的应用

针对扬子石化公司丁二烯精馏塔原控制系统存在的问题, 利用从集散控制系统(DCS)采集的大量现场数据和用机理模型得到的模拟数据, 运用前向反馈(BP)神经网络软测量技术，构造了产品丁二烯和总炔含量的自适应软测量仪表, 设计了一套控制系统. 实际监测数据表明, 这套控制系统可实现产品质量的闭环控制.     

The characteristics of a non-isothermal loop crystallizer

A model for a recirculating draft-tube crystallizer is presented which accounts for spatial variations in environment. The model consists of a well-stirred section to account for evaporative cooling and two plug flow portions in a recirculating mode. The calculations show that the average product crystal size for the recirculating crystallizer is significantly reduced compared to that of a well-stirred crystallizer if the circulation rate is not sufficiently high. The circulation time in most cases needs to be less than one-tenth the crystallizer residence time to ensure well-stirred behavior.     

Modeling of industrial nylon‐6,6 polymerization process in a twin‐screw extruder reactor. II. Neural networks and hybrid models

This article describes the application of neural networks and hybrid models to the finishing stage of nylon‐6,6 polycondensation in a twin‐screw extruder reactor. A planned experiment in the industrial and in the pilot plant was employed to build the neural network and the hybrid model. The hybrid model combines information calculated from the phenomenological model with the neural network model. The comparison of experimental with calculated data shows good agreement. During two years, industrial data were collected. The comparisons of the models' prediction with these data were performed and reasonable results are achieved from the industrial point of view. These models help an increase of industrial production of about 20%. © 1999 John Wiley & Sons, Inc. J Appl Polym Sci 72: 905–912, 1999     

基于模糊神经网络PID控制的污水处理应用研究

针对活性污泥污水处理系统具有复杂的非线性和时变性,传统的控制方法存在着精度不高,自适应能力差等缺点,提出一种模糊神经网络PID控制方法,将模糊神经网络与PID相结合,既发挥了PID控制的优势,又增加了模糊神经网络自学习和处理定量数据的能力,并且其中采用了动态递归神经网络对污水处理系统进行模型辨识。该控制方法能够快速、有效地使曝气池中溶解氧浓度达到期望值,并且具有较好的控制效果与控制精度。仿真结果验证了该控制方法的有效性和正确性。     

Hybrid Neural Network Models Applied to a Free Radical Polymerization Process

This article presents different ways of obtaining hybrid models, which are composed of a simplified phenomenological model and one or several neural networks. As an example, we consider free radical polymerization of methyl methacrylate, achieved through a batch bulk process, in which modeling of conversion and polymerization degrees is analyzed. Kinetics of the process is described through a simplified phenomenological model that does not take into account the gel and glass effects. This last part of the process, which is more difficult to model, is rendered by means of feed-forward neural networks with one or two hidden layers. In the present paper, the hybridization procedure is made in three ways: 1) the neural network corrects the outputs of the simplified kinetic model by modeling the residuals of conversion and polymerization degrees; 2) the neural network provides accurate values of the rate constants to the simplified kinetic model; 3) the neural network models that part of the process in which gel and glass effects appear. It is demonstrated that accurate results are obtained in all three cases, and the hybrid models are easily created and manipulated, especially because they are based on neural networks with quite simple topologies.     

改进的双模型结构RBF神经网络及其应用

提出了离线结构学习和在线权值校正相结合的双模型结构RBF神经网络,以离线学习和在线校正相结合的方式实现网络的自学习和自校正,满足了软测量仪表现场应用的要求。针对应用过程中出现预测误差过大的现象,通过对网络算法进行分析,研究影响网络预测精度的因素,在此基础上,提出了以K均值聚类法和递推下降算法相结合的RBF神经网络建模改进算法,仿真结果和实际应用证明了改进算法的有效性。     

Simultaneous Measurement of Mean Bubble Diameter and Local Gas Holdup Using Ultrasonic Method with Neural Network

One of the greatest challenges in the characterization of bubbles in a bubble column has been the prediction of the bubble diameter and the gas holdup. In this study a novel technique for predicting the mean bubble diameter and the local gas holdup using a non‐invasive ultrasonic method with neural network was investigated. The measurement parameters of the energy attenuation and the transmission time difference of ultrasound are used to obtain the mean bubble diameter and the local gas holdup in an air‐water dispersion system using neural network reconstruction. Bubble size distributions in a 2‐D bubble column are obtained experimentally by using a photographic method. An adequate selection of the neural network structure has been carried out to represent the training data. The representative results using the present structure show good agreement with the measured data.     

基于混合遗传算法的人工神经网络模型及其对有机化合物熔点的预测

为了避免BP神经网络在训练过程中收敛于局部极小的缺陷,采用自适应交叉变异、最优保存的混合遗传算法对BP网络的权值和阈值进行优化,从而提出一种新的基于混合遗传算法的神经网络模型.该算法首先对一给定的网络结构,采用混合自适应交叉变异和最优保存策略,取各自的长处,用尽可能少的搜索代数找到问题的最优解,从而既防止算法陷入局部最优,又保证算法有较好的平均适应值和最佳的适应值个体.采用上述优化策略的人工神经网络可明显改善收敛的稳定性和收敛速度,并确保网络收敛于全局极小点.人工神经网络运用于物性数据的预测是一个具有潜力和有待开发的领域.运用该模型,根据有机化合物的分子量、临界密度、正常沸点和偶极矩,对其熔点进行预测.预测结果表明：提出的混合遗传算法神经网络优于其他算法神经网络,而且预测结果优于文献上已有的Joback方程和许氏方程的计算值.     

Effects of sweating time and cooling strategy on purification of N-vinyl-2-pyrrolidinone using a melt crystallizer

A melt crystallization process is proposed to produce high-purity n-vinyl-2-pyrrolidinone (NVP). To produce high purity products, operation strategy plays key role in the melt crystallizer. We investigated the cooling strategy and optimal sweating time using a batch-type melt crystallizer. A slow cooling followed by a slow heating was found to be an effective temperature profile to produce high purity of NVP. The optimal sweating time was found to be about 20 minutes. For industrial application, a cascade melt crystallizer which consists of four stages was constructed and the proposed crystallization/sweating scheme was applied. Using the new melt crystallizer, NVP more than 99.99% purity can be produced in semi-continuous mode.     

Neural network-based hybrid models developed for free radical polymerization of styrene

In the present work, the free radical polymerization of styrene is modeled by considering the phenomenology of the process (a simplified model, which does not include the diffusional effects, gel, and glass effects) in combination with an empirical model represented by an artificial neural network. Differential evolution (DE) algorithm, belonging to the class of evolutionary algorithms, is applied for developing the neural models in optimal forms. For improving the results—predicted conversion and molecular weights as function of time, temperature, and initiator concentration—different combinations between phenomenological model and neural network are tested; also, individual and stacked neural networks have been developed for the polymerization process. This methodology based on hybrid models, including neural networks aggregated in stacks, provides accurate results.     

Extended kalman filter based nonlinear geometric control of a seeded batch cooling crystallizer

A nonlinear dynamic model of a seeded potash alum batch cooling crystallizer is presented. The model of the batch crystallizer is based on the conservation principles of mass, energy and population. In order to maintain constant supersaturation, a nonlinear geometric feedback controller is implemented. It is shown that compared to a natural and a simplified optimal cooling policies, the nonlinear geometric control (NCC) leads to a substantial improvement of the final crystal quality. An extended Kalman filter (EKF) is used as a closed loop observer for this nonlinear system to predict the non‐measurable state variables. It is found that the EKF is capable of effectively predicting the first four leading moments of the population density function. The effectiveness of the EKF based nonlinear geometric controller in the presence of plant/model mismatch is also studied. Simulation results show that the EKF based nonlinear geometric controller is reasonably robust in the presence of modeling error.     

Application of Model Predictive Control to a Continuous Multiple‐Effect Crystallizer

An infinite horizon model predictive controller (IHMPC) with zone control is applied to a continuous five‐effect evaporative sodium chloride crystallizer. Firstly, a phenomenological dynamic model of the process is developed considering mass, energy, and moment balances coupled to crystallization kinetics. The developed model plays the role of the real system in order to study the proposed optimization/control strategy. The proposed approach is compared to a classical proportional integral derivative (PID) control system. The control strategy based on the prediction of the future state of the plant provides a faster response, a better stability to the process, and a reduction in energy consumption.     

Neural network simulation for non-MSMPR crystallization

A neural network model has been developed for the simulation of steady state industrial crystallizers where, in general, the crystal size distribution cannot be described by simple mass and energy balances, i.e. they are non-MSMPR crystallizers. The model is based on fundamental equations of steady state suspension crystallization. The parameters in the nucleation rate have been chosen for the simulation of different chemicals. The particle size distribution of the product is expressed by the Rosin–Rammler equation. Different operating modes and deviations in crystal size distribution caused by the suspension being imperfectly mixed are presented by different values of modified Rosin–Rammler number. The ranges of variables in the neural network have been chosen based on data for industrial crystallizers. The dominant size of particle, and the productivity of the crystallizer can be predicted with input information. Thus, this neural network can be used for most chemicals and for different kinds of operating conditions. The results predicted with the neural network have been verified by solving the fundamental equations and by comparison with experimental data.     

A HYBRID CFD COMPARTMENTALIZATION MODELING FRAMEWORK FOR THE SCALEUP OF BATCH COOLING CRYSTALLIZATION PROCESSES

Crystallizer design is often hindered by the lack of scaleup rules, hydrodynamic information, and predictive crystallization modeling tools. A hybrid CFD compartmentalization batch cooling crystallizer model is proposed to take into account localized mixing, heat transfer, and fluid hydrodynamics, combined with key process engineering information obtained on a laboratory scale. The compartments were identified using CFD simulations based on the crystallizer geometry and operating conditions. The population, mass, concentration, and energy balance of each compartment is modeled separately as a well-mixed MSMPR unit with input and output streams. The software gPROMS (Process Systems Enterprise Limited) is a process-modeling tool that can facilitate compartmental modeling and will be used for the prediction of the crystallization behavior upon scaleup.     

A HYBRID CFD COMPARTMENTALIZATION MODELING FRAMEWORK FOR THE SCALEUP OF BATCH COOLING CRYSTALLIZATION PROCESSES

Crystallizer design is often hindered by the lack of scaleup rules, hydrodynamic information, and predictive crystallization modeling tools. A hybrid CFD compartmentalization batch cooling crystallizer model is proposed to take into account localized mixing, heat transfer, and fluid hydrodynamics, combined with key process engineering information obtained on a laboratory scale. The compartments were identified using CFD simulations based on the crystallizer geometry and operating conditions. The population, mass, concentration, and energy balance of each compartment is modeled separately as a well-mixed MSMPR unit with input and output streams. The software gPROMS (Process Systems Enterprise Limited) is a process-modeling tool that can facilitate compartmental modeling and will be used for the prediction of the crystallization behavior upon scaleup.     

Use of Neural Network–Ultrasonic Technique for Measuring Gas and Solid Hold‐ups in a Slurry Bubble Column Bubble Colum

A novel technique for measuring simultaneously the gas and solid hold‐ups in a slurry bubble column using a combination of neural network–ultrasonic method was investigated in this study. A one‐dimensional model using the basic parameters of ultrasound (the energy attenuation and the velocity change in terms of the transmission time difference) for measuring the gas and the solid hold‐ups has been proposed to show the complexity of the system. The three layers feed‐forward neural network (3‐FFNN) structure has been used to try and solve the nonlinear relationship between parameter sensing and measurement purpose. An adequate selection of the neural network structure has been chosen to perform the relationship between the measurement sensing (input of the network) and the measurement purpose (output of the network). Preliminary representation results of the gas and the solid hold‐ups using the proposed method compare relatively well with measured data.     

A study of the emotional evaluation models of color patterns based on the adaptive fuzzy system and the neural network

Proposed in this article are two kinds of emotional models based on the neural network and the adaptive fuzzy system that can transform the physical features of a color pattern into its emotional features. The purpose of this system of models is to evaluate the neural network and adaptive fuzzy system for its ability to model psychological experimental data in a way similar to what a human expert would do. Construction of the models was motivated by Soen's psychological experiments, in which he found that such physical features as average hue, saturation, and intensity and the dynamic components of color patterns affected the emotional features represented by a pair of adjectives having opposite meanings. One is based on the neural network in the proposed models, and the other consists of two adaptive fuzzy rule bases and a γ model, a fuzzy set operator, to fuse the evaluation values produced by them. The proposed models showed superior performances compared to Soen's model in the approximation of nonlinear transforms, whereas the latter showed an advantage in obtaining the linguistic interpretation from the trained results. The evaluated results of color patterns can be used to construct a emotion‐based color‐pattern retrieval system, which would be able to recommend the color patterns of a desired human feeling. We believe that in linguistic queries of human feelings, these color‐pattern retrieval systems would be able to select from a gallery the corresponding textile designs, wallpapers, or pictures. © 2002 Wiley Periodicals, Inc. Col Res Appl, 27, 208–216, 2002; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/col.10052