基于选择性集成核学习算法的固废焚烧过程二噁英排放浓度软测量

城市固废焚烧(MSWI)过程排放的二噁英 (DXN)是被称为“世纪之毒”的持续性污染物。该过程的多阶段、多温度区间的物理化学特性导致DXN排放浓度的机理模型难以构建。工业实际中通常以月或季为周期耗时近1周时间在实验室以离线化验方式滞后检测。针对这些问题，提出了基于选择性集成(SEN)核学习算法的DXN排放浓度软测量方法。首先，基于先验知识给出候选核参数集和候选惩罚参数集，采用核学习算法构建基于这些超参数的候选子子模型；然后，耦合优化和加权算法对相同核参数的候选子子模型进行选择与合并，进而得到基于不同核参数的候选SEN子模型集合；最后，再次采用优化和加权算法获得结构与超参数自适应的多层SEN软测量模型。采用UCI平台水泥抗压强度和焚烧过程DXN数据验证了所提方法的有效性。     

基于选择性集成核学习算法的固废焚烧过程二噁英排放浓度软测量

城市固废焚烧(MSWI)过程排放的二噁英(DXN)是被称为"世纪之毒"的持续性污染物。该过程的多阶段、多温度区间的物理化学特性导致DXN排放浓度的机理模型难以构建。工业实际中通常以月或季为周期耗时近1周时间在实验室以离线化验方式滞后检测。针对这些问题,提出了基于选择性集成(SEN)核学习算法的DXN排放浓度软测量方法。首先,基于先验知识给出候选核参数集和候选惩罚参数集,采用核学习算法构建基于这些超参数的候选子子模型;然后,耦合优化和加权算法对相同核参数的候选子子模型进行选择与合并,进而得到基于不同核参数的候选SEN子模型集合;最后,再次采用优化和加权算法获得结构与超参数自适应的多层SEN软测量模型。采用UCI平台水泥抗压强度和焚烧过程DXN数据验证了所提方法的有效性。     

基于LSTM神经网络的干燥含水量预测研究

在PVC干燥中,产品的含水量受到温度、流量等多个参数的影响,鉴于参数数据之间的非线性以及序列间的相关性,传统的时间序列方法和传统的机器学习算法已经不能对未来干燥产品的含水量进行精确的预测。长短期记忆网络（LSTM）作为一种基于深度学习中的循环神经网络（RNN）,它在RNN的基础上增加了输入门、输出门以及遗忘门,可以有效地处理RNN在运行大量数据时可能会带来的数据遗忘等问题,特别适合处理具有时间序列的数据。基于Pytorch深度学习框架构造长短期记忆网络模型,对产品含水量进行预测。结果表明：使用该模型对产品的含水量进行预测,其预测值和真实值的走向非常接近,精准度很高。     

基于尺寸驱动的化工设备CAD装配图的实现

分析了化工设备装配图包含的信息 ,提出用基于尺寸驱动的参数化方法建立图形库和用零部件图拼装法来生成化工设备 CAD装配图 ,并对装配图消隐算法进行了探讨。     

基于时间序列的多岛遗传算法在水泥窑中的应用

研究时间序列在水泥回转窑中的建模问题,水泥回转窑熟料的煅烧是一个涉及物理、化学反应的复杂多变量、多扰动非线性过程.针对最小二乘在多参数高阶时间序列模型中难以寻找到较好的系数组,从而不能建立较精确的时间序列模型的情况,采用传统遗传算法也很难寻找出最优参数组,当利用多岛遗传算法对其进行参数寻优时,只要适当地设置算法里的一些关键参数即能得出满意的组解.实验结果表明：经过多岛遗传算法寻优得出的参数建立的时间序列模型,其拟合度满足模型曲线的精度要求.     

循环流化床锅炉热效率在线计算与影响因素分析

提出一种基于正平衡方法的热效率在线计算方法:通过在线采集锅炉运行中主蒸汽、给水、一二次风等锅炉运行参数,计算锅炉有效利用能量和输入能量,实现热效率的在线计算;采用福建某135 t/h工业CFB锅炉发电机组现场数据进行仿真计算,仿真结果验证了算法的有效性.在此基础上采用正交试验设计方法分析影响CFB锅炉热效率的影响因素;由于工业现场采样的操作数据难以得到各个因素不同水平下的热效率的对应值,故首先采用工业数据和热效率计算数据建立基于LM-BP算法的人工神经网络热效率预测模型.根据模型预测数据设计正交试验,结果表明对于该锅炉风煤比、主汽压力和炉床温度对该锅炉热效率的影响最大.最后进行工业调优试验,试验结果验证了分析结果并初步实现了锅炉热效率的在线优化.     

多频率系统动态插值神经网络软测量建模

针对某些化工过程关键变量难以在线测量的问题,提出了一种基于多采样率系统的时间序列神经网络的软测量建模方法,建立了动态插值神经网络模型,并利用增强粒子群算法实现了网络参数的优化。将此方法用于实验室模拟建模,实现了变量的在线预估,并对网络的训练效果和泛化性能进行了分析,表明其建模效果明显优于普通静态神经网络。     

增塑剂TP-95和TP-90B对丁腈橡胶性能的影响

研究了增塑剂TP-95和TP-90B对丁腈橡胶(NBR)低温性能、硫化特性、力学性能和老化性能的影响.并与常用增塑剂DOA,DOP进行对比.结果表明,添加含醚基的TP-90B和含醚酯基的TP-95使NBR硫化胶的玻璃化温度明显降低,低温拉伸性能和低温脆性性能优于DOA和DOP.TP-90B和TP-95对胶料硫化有一定的促进作用.硫化胶的100%和300%定伸应力较高.TP-95在热空气老化条件下具有良好的热稳定性,而TP-90B较易挥发.     

改进的Prony算法在频谱估计中的应用

在对Prony算法分析与研究的基础上,将基于奇异值分解的总体最小二乘法引入Prony算法,并将该改进的Prony算法应用于频谱估计.改进的算法弥补了信号序列在伴随噪声干扰和数据序列较短的情况下,快速傅立叶变换的方法无法对信号频率进行有效提取的缺点,最后通过实验仿真证明了该改进的Prony算法在频谱估计中应用的有效性.     

乙炔加氢反应过程混合建模与优化

针对传统单一建模方法所构建的乙炔加氢反应器数学模型存在预测性能无法满足工业实际应用需求的问题,提出了一种机理与神经网络嵌套的建模方法,充分利用机理模型包含的能质约束信息降低神经网络模型的约束违反度,得到了能够良好描述实际工业乙炔加氢反应过程特性的混合模型。基于反应器混合模型,研究了以运行效益为目标函数的优化问题。主要决策变量包括：一段反应器进料中氢气与乙炔的摩尔比（R _H/A）、进料温度和反应器运行周期等几个关键参数。针对反应器长期运行后,催化剂活性降低造成的处理能力下降的问题,提出了反应温度补偿机制和R _H/A并行调节的运行优化策略,并采用序列法对反应器运行周期进行离散化处理。通过引入差异化变异策略、潜在解替代策略对两阶段差分算法进行改进,采用增量式编码法结合改进两阶段差分算法,对优化问题进行求解。结果证实了优化策略与改进算法的有效性,并据此确定了反应器最佳运行方案。     

A Graph-based Ant Colony Optimization Approach for Integrated Process Planning and Scheduling

This paper considers an ant colony optimization algorithm based on AND/OR graph for integrated process planning and scheduling (IPPS). Generally, the process planning and scheduling are studied separately. Due to the complexity of manufacturing system, IPPS combining both process planning and scheduling can depict the real situation of a manufacturing system. The IPPS is represented on AND/OR graph consisting of nodes, and undirected and directed arcs. The nodes denote operations of jobs, and undirected/directed arcs denote possible visiting path among the nodes. Ant colony goes through the necessary nodes on the graph fromthe starting node to the end node to obtain the optimal solution with the objective of minimizing makespan. In order to avoid local convergence and lowconvergence, some improved strategy is incorporated in the standard ant colony optimization algorithm. Extensive computational experiments are carried out to study the influence of various parameters on the system performance.     

Parallel observability analysis on networks of workstations

In this work we present the parallelisation of the global strategy with first least-connected node (GS-FLCN), which is a novel structural technique for the classification of unmeasured variables in process plant instrumentation design. The algorithm aims at partitioning the process’ occurrence matrix to a specific block lower-triangular form. A parallel master–workers philosophy is employed to search for all the paths of a given length existing in the associated graph. The code was conceived for distributed environments and the implementation was carried out using the parallel virtual machine (PVM) library. The performance of the parallel algorithm was tested for industrial case studies and the results were compared with those yielded by the sequential version. The time savings achieved thanks to the parallelisation were significant. Besides, in the parallel version, more paths can be explored per unit time. In practice, this implies greater robustness.     

An improved algorithm for diagnosis of system failures in the chemical process

The detailed techniques for improving the efficiency of a fault-diagnosis algorithm can be based on the signed directed graph. This graph represents the structure of the system, and the pattern on the graph represents a state of the system. The usefulness of the improved algorithm is demonstrated using a model of a chemical process comprising a reactor, a heat exchanger and a distillation tower.     

A graphical approach to cause and effect analysis of chemical processing systems

A method for cause and effect analysis of chemical processing systems has been developed on the concept of a signed directed graph. The signed directed graph, called a “model graph,” is used for representing cause and effect relationships among process variables in a system, and a “Pattern” on the graph is introduced to represent a state of the system, which is specified by “0” for normal state, “ + ” and “ ? ” for larger and smaller deviation from the normal state, respectively. For a given model graph and pattern on it, the cause of state changes and the manner of their propagation are easily found out by the present approach, which is mainly based on the depth-first search algorithm developed by Tarjan. As an illustrative example, a propane vaporizer has been taken to show the usefulness of the present approach.     

Improved infeasible path optimization for sequential modular simulators—I: The interface

Recently, it was shown that chemical processes modeled by steady-state simulators could be optimized without repeatedly converging the process simulation. Instead, optimization and simulation of the process can be performed simultaneously (along an infeasible path), thus leading to much more efficient performance. In this two-part study, several improvements to this infeasible path approach are described. This first paper deals with improvements to the interface between the optimization algorithm and the process simulator. Here one is primarily concerned with obtaining the necessary function and gradient information for the optimization with minimum computational effort from the process simulator. The architecture of sequential modular simulators, the structure of process optimization problems and any sources of error in obtaining the necessary gradient information for the optimization algorithm are considered. To this end, a chainruling algorithm is derived that allows the incorporation of analytic derivative information for parts of the flowsheet and generally leads to less frequent evaluation of the flowsheet modules. This algorithm is demonstrated on three process optimization problems. The results indicate significant improvement in performance.     

EXPERIENCE WITH ADOPTING ONE-PARAMETER IMBEDDING METHODS TOWARD CALCULATION OF COUNTERCURRENT SEPARATION PROCESSES

The one-parameter imbedding method (also called homotopy or continuation) was adopted toward solution of large sets of nonlinear algebraical equations describing counter-current separation processes. Different imbedding functions were tested on a spectrum of difficult distillation problems ranging from distillation of hydrocarbons to strongly nonideal distillation problems. For the one-parameter imbedding functions studied in this report the classical Newton-Raphson Formula can be easily generated after an appropriate selection of the control parameters. Two different approaches were used to solve the homotopy equations: i) marching integration, ii) sequential use of the Newton-Raphson method. The one-parameter imbedding technique represents a trade-off between robustness and computation time. The algorithm is more robust than the Newton-Raphson technique, however, the computational time is usually higher. A combination of the one-parameter imbedding and the Newton-Raphson approach seems to be a very efficient method, the solution is approached by the one-parameter imbedding technique and the Newton-Raphson method is used to finish the iteration process. Geometrical interpretation of convergence is presented.     

基于减法聚类的焦炉集气管压力操作模式提取及迁移重构

基于集气管压力控制特点和生产过程中产生的大量历史数据和以往操作经验的基础上，提出一种基于减法聚类的焦炉集气管压力控制的操作模式提取及迁移重构方法。基于减法聚类算法进行操作模式的提取，最终形成操作模式优化数据库，对压力设定值进行优化，采用基于模型迁移思想的模式重构策略实现操作模式的修正。工业应用试验验证了所提方法的有效性。     

Application of stepwise regression for dynamic parameter estimation

Dynamic parameter estimation in cases where it may be impossible to identify all the model parameters is considered. The objective is to obtain reliable estimates to the maximal number of physical parameters in a stable regression model where the modeling of the noise in the data is avoided. The modifications required in the stepwise regression algorithm to accommodate various nonlinear terms in the regression model are investigated and a new algorithm is presented. The algorithm considers the hierarchy among the parameters, the initial trends of the experimental data curves and the initial values of the state variables in order to establish a minimal initial set of parameters to be included in the model. Additional parameters are then added in a stepwise manner, while considering the hierarchy of the parameters and the associated reduction of the objective function value. The process continues as long as significant and physically feasible values for the parameters are obtained. The new method is demonstrated with several examples from the literature. Additional issues investigated include the proper combination of the simultaneous and sequential solution methods in the stepwise regression algorithm, the preferred method for the estimation of the derivatives and the effect of variable scaling.     

A sequential sampling plan for exponential distribution

In this article, a sequential variable sampling plan is studied. Suppose that the quality of an item in a batch is measured by a random variable with exponential distribution; its parameter is unknown having a gamma prior distribution. Then by using Bayesian approach and considering a Markov decision process, the optimality equations for the minimum total expected cost are formulated. We show that an optimal decision rule will have a control limit structure and monotonicity. A backward induction method is suggested that is a finite algorithm for the numerical solution of the sequential sampling plan.     

Extracting dynamic features with switching models for process data analytics and application in soft sensing

In recent decades, soft sensors have been profoundly studied and successfully applied to predict critical process variables in real‐time. While dealing with various application scenarios, data‐driven methods with representation learning possess great potentials. Latent features are formulated in these approaches to predict outputs from correlated input variables. In this study, a probabilistic framework of feature extraction is proposed in the context of process data analysis. To address switching behaviors in industrial processes, multiple emission models are utilized to construct latent space. To address temporal correlations from continuously operating processes, a dynamic model is implemented in latent space. Bayesian learning strategy is then developed for parameters estimation, where modeling preferences and uncertainties from multiple models are considered. The effectiveness and practicability of the proposed feature extraction algorithm are illustrated through numerical simulations, as well as an industrial case study. © 2018 American Institute of Chemical Engineers AIChE J, 64: 2037–2051, 2018