递归T-S模糊模型的神经网络

在常规T-S模糊神经网络的基础上加入动态递归元件,提出了递归T-S模糊模型的神经网络。在系统辨识中采用无监督聚类算法和动态反向传播算法训练该递归神经网络的参数,给出了该递归网络的逼近性证明。辨识效果与常规T-S模糊模型作比较,说明递归T-S模糊模型的神经网络在非线性系统辨识中表现出更好的性能。     

一种扰动自适应的鲁棒预测控制算法

针对实际生产中扰动的时变性,提出了一种扰动自适应的鲁棒预测控制（RAMPC）算法以提高扰动抑制性能。采用时间序列（ARMA）模型在线辨识系统的不可测扰动,通过基于多次迭代思想的递推辨识算法（multi-iteration pseudo-linear regression,MIPLR）来保证在线辨识的质量和收敛速度。考虑到数据与辨识模型的不确定性,改用min-max形式描述MPC算法的控制作用优化命题,并将在线辨识过程中的误差数据引入min-max命题,使在线辨识与控制作用鲁棒优化求解紧密结合起来,提高算法鲁棒性。进一步将此min-max问题转换为一个等效的非线性min问题,并采用多步线性化方法实现快速求解,解决了传统min-max方法在线计算负荷高的问题。仿真结果表明了该算法的有效性。     

一种参数辨识与模型匹配相结合的自适应控制

讨论了对象参数辨识与模型匹配控制(MMC)相结合的自适应控制方法。该方法是基于RIV 在线参数辨识求取时变对象参数,进而利用 MMC 方法不断修正反馈与前馈增益,使对象得到与预选模型相匹配的动态响应。实验室一模拟对象与一实际工业对象的数字仿真结果说明了该方法的可行性。该算法能被改进为一种实时在线的计算机控制策略而加以实施。     

基于广义预测控制的间歇生产迭代优化控制

针对间歇生产,提出了一种基于广义预测控制的批次迭代优化控制策略--BGPC,在间歇过程中引入批次间优化的思想,将迭代学习控制ILC和广义预测控制GPC相结合,在GPC实时结构参数辨识的基础上利用前面批次的模型预测误差修正当前批次的模型预测值.该算法能够有效地克服模型失配、扰动和系统参数变化等情况.文章最后以一个数值例子和间歇反应器为对象进行仿真试验,验证了该算法是有效的.     

一类化工过程多变量系统的自适应非线性预测控制

针对化工过程的一类多变量非线性系统,提出了一种自适应非线性预测控制(ANMPC)算法。在采用递归最小二乘法进行预测模型参数在线辨识的基础上,将系统的静态非线性关系用一个反向传播(BP)神经网络稳态模型来表示,通过稳态模型求得的动态增益来进一步校正预测模型的参数。详述了ANMPC控制器设计步骤,通过在一个多变量pH中和过程中的仿真验证了本算法的可行性和有效性。     

基于ROLS算法的递归RBF神经网络结构设计

针对递归RBF神经网络结构难以自适应问题,提出一种基于递归正交最小二乘（recursive orthogonal least squares,ROLS）算法的结构设计方法。首先,利用ROLS算法来计算隐含层神经元的独立贡献度和损失函数,以此判断增加或归为不活跃组的神经元,同时调整神经网络的拓扑结构,并且利用奇异值分解（singular value decomposition,SVD）决定最佳的隐含层神经元个数,以此来删除不活跃组中相对不活跃的神经元,有效地解决了递归RBF神经网络结构冗余和难以自适应问题。其次,利用梯度下降算法更新递归RBF神经网络的参数来保证神经网络的精度。最后,通过对Mackey-Glass时间序列预测、非线性系统辨识和污水处理过程中关键水质参数动态建模,证明了该结构设计方法的可行性和有效性。     

变时滞系统的辨识自适应控制算法的改进与仿真研究

城市供水出水浊度过程控制是大时滞难控对象,Smith预估控制方法的研究有效地改善了时滞带来的控制困难的问题,但是当系统模型不精确时,很难获得好的控制品质。采用时滞辨识自适应方法,有好的控制效果,但当被控系统的时滞参数是时变时,在一定范围内可能存在的时滞辨识的效果不理想,所以提出采用变搜寻区域的辨识方法,可以实时地搜索出真实过程的时滞。该算法用于变时滞的系统,具有良好的控制品质和较强的自适应能力。     

基于ROLS算法的递归RBF神经网络结构设计

针对递归RBF神经网络结构难以自适应问题,提出一种基于递归正交最小二乘(recursive orthogonal least squares,ROLS)算法的结构设计方法。首先,利用ROLS算法来计算隐含层神经元的独立贡献度和损失函数,以此判断增加或归为不活跃组的神经元,同时调整神经网络的拓扑结构,并且利用奇异值分解(singular value decomposition,SVD)决定最佳的隐含层神经元个数,以此来删除不活跃组中相对不活跃的神经元,有效地解决了递归RBF神经网络结构冗余和难以自适应问题。其次,利用梯度下降算法更新递归RBF神经网络的参数来保证神经网络的精度。最后,通过对Mackey-Glass时间序列预测、非线性系统辨识和污水处理过程中关键水质参数动态建模,证明了该结构设计方法的可行性和有效性。     

基于自适应模糊推理的非线性系统辨识器设计

针对传统模糊建模方法中模型参数都是根据经验选取的局限性,提出一种类高斯隶属函数,推导了基于类高斯隶属函数的自适应模糊推理模型,利用Stone-Weierstrass定理证明了该模型能以任意精度逼近非线性系统.将自适应模糊推理模型应用于非线性动态系统辨识中,设计了非线性系统辨识器,采用梯度下降算法学习模型中参数,通过仿真得到了较好的辨识效果.     

基于递归模糊神经网络的污水处理控制方法

针对污水处理过程具有非线性、大时变等问题,提出了一种基于递归模糊神经网络的多变量控制方法。该方法通过递归模糊神经网络控制器自适应地获得对操作变量的控制精度,控制器在常规BP学习算法的基础上采用学习率自适应学习算法且引入了动量项来训练网络参数,避免网络陷入局部最优,提高了网络对系统的控制精度。最后,基于仿真基准模型(BSM1)平台对第五分区中的溶解氧和第二分区中的硝态氮控制进行动态仿真实验,结果表明,与PID、前馈神经网络和常规递归神经网络相比,该方法能有效提高系统的自适应控制精度。     

塑料注射充模流动分析的自适应隐式方法

提出了一个注射充模数值模拟隐式方法的时间步长选用标准,依据“一时间步,一单元层”的思想,采用最小二乘递推法估算每一步的时间步长,并依据压力估算每个控制体积的半注满时间,从而可以重构任何时间的流动前锋。计算表明本方法极大地提高了隐式方法的适用性,并把注射充模流动数值模拟的计算时间复杂度降到结点数的平方。     

注塑流动与传热分析的自适应隐式控制体积法

在注塑流动与传热分析中采用自适应隐式控制体积法,利用自动控制的参数识别方法来自动调节时间步长,在每个时间步长流动将向前推进大约半个单元大小的距离,厚度方向的温度采用契比雪夫配点法进行计算,压力方程的求解方法是预条件共轭梯度法。计算结果表明,自适应隐式控制体积法的时间复杂度是结点数目的平方,而且1h之内可以求解10000结点规模的问题。     

ADAPTIVE POLE-PLACEMENT CONTROL OF A CONTINUOUS POLYMERIZATION REACTOR

A simulation of the polymerization of methylmethacrylate in a CSTR is adaptively controlled by two types of pole-placement algorithms. The strongly nonlinear polymerization process, exhibiting multiple steady states, presents difficult control problems for conventional feedback controllers. The performance of an adaptive explicit SISO controller and that of an adaptive implicit multivariable controller are compared and evaluated as applied to this process. The plant is identified by a recursive least squares estimator. Modifications made to the estimation algorithm help to maintain adequate closed-loop results. A simple warm-up procedure is introduced that successfully initializes the controller and estimator during plant start-up. Good servo and regulatory control are achieved by both pole-placement schemes.     

递归核PCA及其在非线性过程自适应监控中的应用

PCA、PLS作为常用的多变量统计监控算法,一般适用于线性、定常的过程。针对实际工业过程的时变、非线性特性,提出了一种递归核PCA（RKPCA）方法用于非线性过程的自适应监控。RKPCA算法通过将递归奇异值分解推广到核空间,给出了核形式描述的递归KPCA算法,运算复杂度比KPCA明显降低,保证非线性监控模型能够在线更新。在Alstom工业燃气发生装置上的自适应监控表明,所提出的RKPCA算法能够及时跟踪非线性过程的时变特征,保证了监控模型的有效性。     

基于主元相似度的间歇过程操作曲线递推优化

针对一类较长周期的间歇过程操作优化问题，提出了一种基于正常运行批次的数据驱动型操作曲线递推优化方法。首先采用分段离散化方法将原非线性优化问题转化为线性优化问题，再利用主元分析对离散化后的高维时段变量进行降维处理，然后在降维后的主元平面中，基于时段变量与最终产品指标间的余弦相似度，实现对原操作曲线的摄动优化。考虑到时段变量方差和相似度随批次会发生变化，建立了递推算法以实现操作曲线的递推更新。最后将该方法应用于某化工产品的间歇结晶过程中，结果验证了所提方法的有效性。     

基于稳定Hammerstein模型的在线软测量建模方法及应用

针对复杂工业过程中由于存在未建模动态和不确定干扰,导致关键变量的软测量精度下降的问题,提出了一种基于稳定Hammerstein模型(H模型)的在线软测量建模方法。H模型的非线性增益采用带有时变稳定学习算法的小波神经网络模型,线性系统部分采用基于递推最小二乘的ARX模型,基于输入到状态稳定性理论证明了H模型辨识误差的有界性。其中小波神经网络具有表征强非线性的特性,稳定学习算法可抑制未建模动态和不确定干扰的影响,改善了模型的预测精度和自适应能力。以典型非线性系统和实际污水处理过程为例进行了仿真研究,结果表明,基于稳定H模型的软测量方法具有较高的在线软测量精度。     

基于ARX-PLS解耦框架下的多回路约束迭代模型预测控制方法(英文)

A multi-loop constrained model predictive control scheme based on autoregressive exogenous-partial least squares （ARX-PLS） framework is proposed to tackle the high dimension, coupled and constraints problems in industry processes due to safety limitation, environmental regulations, consumer specifications and physical restric-tion. ARX-PLS decoupling character enables to turn the multivariable model predictive control （MPC） controller design in original space into the multi-loop single input single output （SISO） MPC controllers design in latent space. An idea of iterative method is applied to decouple the constraints latent variables in PLS framework and recursive least square is introduced to identify ARX-PLS model. This algorithm is applied to a non-square simulation system and a stirred reactor for ethylene polymerizations comparing with adaptive internal model control （IMC） method based on ARX-PLS framework. Its application has shown that this method outperforms adaptive IMC method based on ARX-PLS framework to some extent.     

ADAPTIVE MULTIVARIABLE PID CONTROLOF A DISTILLATION COLUMN WITH UNKNOWN AND VARYING DEAD TIME

A new, reliable, and easy-to-use adaptive control strategy has been developed to overcome the long-existing difficulties in adaptive control practice caused by unknown and varying process dead time. A self-tuning PID control algorithm is adopted to control a distillation column possessing second-order-plus-dead-time dynamics. The self-tuning strategy is based on recursive least-squares estimation of process parameters. U-D factorization is applied to stabilize the parameter estimation calculations. A variable forgetting factor is used to alleviate wind-up in the estimator. A simulation study and an experimental evaluation demonstrate the capability of the adaptive algorithm.     

APPLICATION OF MULTIVARIABLE NONLINEAR ADAPTIVE GENERIC MODEL CONTROL TO A PACKED DISTILLATION COLUMN

Nonlinear adaptive generic model control and self-tuning PID control systems were applied to control the top and bottom product temperature of a packed distillation column separating methanol-water mixture. In the first control algorithm, an adaptive generic model control (AGMC) structure was proposed for dual temperature control of the system. In the second control algorithm, nonlinear self tuning PID (NLSTPID) control based on pole-placement technique was used to control the same system. For NLSTPID control purposes pseudo random binary sequence (PRBS) signal and recursive identification algorithm were used to estimate the relevant parameters of a polynomial NARMAX model. In this work, real-time application has been carried out. In both dynamic and control studies, perturbations in feed composition were utilized as the disturbance, and the reboiler heat duty and the reflux ratio were selected as the manipulated variables. The control performances have been obtained by using ISE and, in general, AGMC results were better than those of the STPID control algorithm.     

Application of generalized predictive control to baker's yeast production

Generalized Predictive Control (GPC) was applied in the production of baker's yeast. The bioreactor was modeled with the autoregressive integrated moving average exogenous (ARIMAX) parametric difference equation model.A 2 L bioreactor with a cooling jacket was used for collecting input-output data. In order to measure pH, temperature, and dissolved oxygen in the bioreactor growth medium, suitable sensors were placed in the bioreactor. Medium temperature and the heat of the immersed heater were selected as output and manipulated variable, respectively. Square wave and a pseudo-random binary sequence (PRBS) signal were used as disturbance. Model parameters were calculated by using the recursive least square parameter estimation method. Bioreactor temperature was controlled theoretically using the GPC algorithm. The control performance was investigated by giving positive and negative step responses to the set point. The GPC algorithm holds the bioreactor temperature succesfully at the optimal set point. Optimum values of the maximum costing horizon ( N 2 ), control horizon ( N U ), and control weighting ( u ) were found to be 10, 1, and 0.005, respectively.