基于典型相关分析和数据自回归处理的BP神经网络在聚丙烯熔融指数预报中的应用

由于聚丙烯生产是一个大量参数相互耦合的强非线性过程,使得传统的机理建模受到一定的限制。提出基于典型相关分析和数据自回归处理的BP神经网络软测量建模,通过可测变量来推知聚丙烯熔融指数。应用典型相关分析选择与输出熔融指数关系较大的独立输入变量,数据自回归处理校正一系列带有误差的量测数据,而BP神经网络用来刻画过程的非线性特征。最后,将提出的算法应用到聚丙烯大型生产工艺中进行熔融指数的预报建模并进行实例仿真,仿真结果表明该算法有较强的建模精度。     

基于操作域划分的聚丙烯熔融指数软测量

讨论了如何建立聚丙烯熔融指数软测量模型及模型更新问题.首先根据聚丙烯反应器中的氢气浓度划分操作域,对于每个操作域,用一种新的非线性部分最小二乘方法建立熔融指数软测量子模型,然后将各个子模型进行组合,建立全局模型.为了使模型适应过程的变化,提出一种递推非线性部分最小二乘算法,利用新获得的数据对原模型进行更新.同时基于滑动窗方法,提出模型在线估计和更新策略.实际应用结果表明,模型取得了很好的估计性能,计算精度满足工业生产的实际要求.     

基于AKPLS方法的聚丙烯熔融指数软测量

针对聚丙烯装置熔融指数软测量中的非线性和多牌号切换问题,提出一种基于自适应核偏最小二乘（AKPLS）的软测量方法。通过对聚丙烯装置反应系统进行机理分析,采用非线性PLS——KPIs方法来拟合辅助变量和熔融指数之间的函数关系。为适应装置生产多牌号产品的现状,进一步提出KPLS自适应策略,以软测量模型的泛化误差作为优化目标,对KPLS模型系数进行在线更新。工业数据应用结果表明,所提出的AKPLS方法能够比PLS、KPLS方法更准确地预报聚丙烯熔融指数的变化。     

基于GA优化的非线性岭回归方法及其在软测量中的应用

提出一种基于遗传算法的非线性岭回归建模方法(GA-NLRR).该方法的核心是先通过RBF的转换实现输入样本的非线性映射,然后用岭回归方法进行线性建模,并采用遗传算法优化岭参数k.该建模方法能很好弥补常规岭回归方法的不足,即无法处理复杂非线性问题和岭参数难确定的问题.将该方法应用于溶剂脱水塔的软测量建模中,仿真研究表明:使用GA-NLRR建立的模型具有很好的预测精度.     

基于DBN-ELM的聚丙烯熔融指数的软测量

针对聚丙烯熔融指数软测量中预测精度不高的缺点,将基于深度置信网络-极限学习机（DBN-ELM）的软测量方法应用到熔融指数的软测量中。与传统深度置信网络（DBN）不同的是,该方法将极限学习机（ELM）算法运用到深度置信网络的训练中。首先用深度置信网络对原始数据进行数值分析来提取特征,然后将提取的特征输入到极限学习机中进行训练,得到软测量模型。实验验证表明,与支持向量机和单纯的深度置信网络模型相比,该方法具有更高的测量精度。     

基于递推PLS核算法的软测量在线学习方法

针对过程的动态时变特性,提出一种基于PLS核算法的软测量在线学习方法。该方法利用PLS核算法,通过递推学习具有代表性的新样本来改善模型的适应能力,较NIPALS算法具有更高的计算效率;并采用一种同时考虑输入和输出信息的相似度准则,有选择地删除一个或多个冗余样本,更有效地构建了训练样本集。工业聚丙烯熔融指数的软测量建模研究表明,本文提出的方法能够快速有效地跟踪牌号切换中熔融指数的变化。     

基于DBN-ELM的聚丙烯熔融指数的软测量

针对聚丙烯熔融指数软测量中预测精度不高的缺点,将基于深度置信网络-极限学习机(DBN-ELM)的软测量方法应用到熔融指数的软测量中。与传统深度置信网络(DBN)不同的是,该方法将极限学习机(ELM)算法运用到深度置信网络的训练中。首先用深度置信网络对原始数据进行数值分析来提取特征,然后将提取的特征输入到极限学习机中进行训练,得到软测量模型。实验验证表明,与支持向量机和单纯的深度置信网络模型相比,该方法具有更高的测量精度。     

Spheripol工艺丙烯聚合质量模型

针对Spheripol工艺,根据Z-N催化剂聚合机理,对不同牌号聚丙烯产品(包括均聚聚丙烯、无规共聚聚丙烯和抗冲聚丙烯)分别建立了熔融指数和乙烯质量分率的预测模型.并应用实际生产数据,对模型进行优化拟合,得到了模型参数.结果表明,熔融指数和乙烯质量分率预测模型的计算值与分析值之间的平均相对误差分别为3.86%和13.5%,模型可以较为准确地预测树脂质量,对聚合产品质量的软测量及牌号切换的研究有很大帮助.     

嵌入岭回归的BP算法及其在软测量中的应用

针对三层神经网络(ANN)最佳隐节点个数难以确定和随着隐节点个数增加ANN模型易出现过拟合等缺点,提出了嵌入岭回归(RR)的误差反传算法(BP).BP-RR根据样本规模自适应确定隐节点个数,并通过BP算法充分提取样本数据信息.然后,针对隐含层输出可能存在的复共线性,采用RR以预测性能为指标,通过进化算法确定最佳岭参数,进而重新确定隐含层与输出层之间最佳的权值和阈值,克服ANN过拟合,建立具有良好预测性能的模型.将BP-RR应用于建立石脑油干点软测量,结果显示,BP-RR模型具有良好的预测性能.与ANN相比,BP-RR模型鲁棒性强,预测精度高.     

聚丙烯熔融指数软测量

采用机理和统计两种建模方法，建立聚丙烯生产过程熔融指数软测量模型。根据聚合反应机理和反应动力学，建立丙烯聚合过程的机理模型。统计模型则采用神经网络／部分最小二乘（NNPLS）方法。现场投运结果表明，模型计算速度和精度能满足工业现场在线实时运行的要求。     

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.     

Batch-to-batch optimal control of a batch polymerisation process based on stacked neural network models

A neural network based batch-to-batch optimal control strategy is proposed in this paper. In order to overcome the difficulty in developing mechanistic models for batch processes, stacked neural network models are developed from process operational data. Stacked neural networks have enhanced model generalisation capability and can also provide model prediction confidence bounds. However, the optimal control policy calculated based on a neural network model may not be optimal when applied to the true process due to model plant mismatches and the presence of unknown disturbances. Due to the repetitive nature of batch processes, it is possible to improve the operation of the next batch using the information of the current and previous batch runs. A batch-to-batch optimal control strategy based on the linearisation of stacked neural network model is proposed in this paper. Applications to a simulated batch polymerisation reactor demonstrate that the proposed method can improve process performance from batch to batch in the presence of model plant mismatches and unknown disturbances.     

遗传Marquardt神经网络识别油气水多相流流型研究

目前利用最优化算法中的Marquardt法改进BP神经网络正受到越来越多的人们的注意 ,但该方法的网络初始权值是随机选取。由于初始权值选取不当将对整个网络的性能产生严重影响 ,因此提出将遗传算法与Marquardt法结合 ,先利用遗传算法全局随机搜索寻优的特性来寻找网络最佳初始权值 ,再用Marquardt法使网络权系数稳定收敛 ,同时应用该方法对油气水多相流流型进行智能识别 ,结果表明该方法能有效学习模式样本 ,学习稳定 ,推广能力强 ,适合于在流型识别等神经网络为中小规模的场合下应用。     

Prediction of the pH and the temperature-dependent swelling behavior of -alginate hydrogels by artificial neural networks

This paper considers the possibility of using artificial neural network models to identify model for swelling behavior as new techniques. Multi-layer feed-forward, radial basis function and generalized regression neural network models were employed to predict the swelling behaviors of Ca²⁺-alginate hydrogels under different environmental conditions of pH and temperature. The results show that an excellent correlation between the experimental and predicted swelling ratios was obtained by the artificial neural networks. Generalized regression neural network has a better performance than the other neural network models. The absolute mean error, the determination coefficient and the standard error of prediction were used as performance criteria. In addition, the performances of the neural network models are significantly superior compared with those of second-order swelling kinetics, quadratic and cubic models of response surface methodology.     

水泥煅烧工段机理分析与神经网络预测方法

采用神经网络的方法建立水泥预分解窑煅烧工段的预测模型。选择合理的状态与控制变量,通过采集实际运行数据来训练神经网络。构建的基于BPNN神经网络的煅烧预测模型能够较好地拟合采样数据,具有较好的泛化能力。     

Optimal control of batch processes using particle swam optimisation with stacked neural network models

An optimal control strategy for batch processes using particle swam optimisation (PSO) and stacked neural networks is presented in this paper. Stacked neural network models are developed form historical process operation data. Stacked neural networks are used to improve model generalisation capability, as well as provide model prediction confidence bounds. In order to improve the reliability of the calculated optimal control policy, an additional term is introduced in the optimisation objective function to penalize wide model prediction confidence bounds. The optimisation problem is solved using PSO, which can cope with multiple local minima and could generally find the global minimum. Application to a simulated fed-batch process demonstrates that the proposed technique is very effective.     

死端微滤酵母悬浮液比阻的预测

本文首先设计了三因素四水平的正交实验表作为建模样本,其次利用人工神经网络方法和多元线性回归方法分别建立了基于操作条件（压力△P=0.04-0.12 MPa,浓度C = 0.3-2.0 g.L^-1,温度T = 20-40℃）的比阻预测模型,以期用于死端微滤过程操作条件的优化,最后以检验样本的相对误差作为衡量指标,分别采用BP人工神经网络方法和多元线性回归方法对死端微滤过滤酵母悬浮液时的比阻进行了预测。研究结果表明：(1) 在本实验范围内,BP人工神经网络模型的最佳拓朴结构为3-7-1,隐层神经元个数为7,学习速率为0.05,学习函数为traingdx, 传递函数为Logsig;用多元线性回归方法得到的比阻与操作条件之间的数学关系式为1.639883+44.2 +0.86217 -0.0607 ; (2)利用BP人工神经网络和多元线性回归方法预测死端微滤比阻的平均相对误差分别为3.55％和5.16％.由此可见,这两种方法都可用于死端微滤比阻预测,并且前者优于后者。     

Deep chemometrics using one-dimensional convolutional neural networks for predicting crude oil properties from FTIR spectral data

The determination of physicochemical properties of crude oils is a very important and time-intensive process that needs elaborate laboratory procedures. Over the last few decades, several correlations have been developed to estimate these properties, but they have been very limited in their scope and range. In recent years, methods based on spectral data analysis have been shown to be very promising in characterizing petroleum crude. In this work, the physicochemical properties of crude oils using Fourier transform infrared (FTIR) spectrums are predicted. A total of 107 samples of FTIR spectral data consisting of 6840 wavenumbers is used. One dimensional convolutional neural networks (CNNs) were used employing FTIR spectral data as the one-dimensional input and Keras and TensorFlow were used for model building. The Root Mean Square Error decreased from 160 to around 60 for viscosity when compared to previous machine learning methods like partial least squares (PLS), principal component regression (PCR), and partial least squares regression with genetic algorithm (PLS-GA) on the same data. The important hyper-parameters of the CNN were optimized. In addition, a comparison of results obtained with different neural network architectures is presented. Some common preprocessing techniques were also tested on the spectral data to determine their impact on model performance. To increase interpretability, the intermediate neural network layers were analyzed to reveal what the convolutions represented, and sensitivity analysis was done to gather key insights about the wavenumbers that were the most important for prediction of the crude oil properties using the neural network.     

Neural network prediction of cetane number and density of diesel fuel from its chemical composition determined by LC and GC-MS

Neural networks were used to correlate and predict the cetane number and the density of diesel fuel from its chemical composition. Cetane number (CN) and density were correlated with 12 hydrocarbon groups in diesel fuel determined by liquid chromatography (LC) and gas chromatography-mass spectrometry (GC-MS). In total, 69 diesel fuels were available for this study: 48 diesel fuels were included in the training data set and 21 in the test data set. Various neural network architectures were trained using the training data set, and the accuracy of the model obtained was examined by using the test data set. For correlating both CN and density in this study, the best neural network architecture was a general regression neural network (GRNN). With the test data set, the mean absolute errors were 1.23 (CN) and 0.002 g/cm³ for the CN and density, respectively. Predictive equations for CN and density of diesel fuel from its chemical composition were also developed with a standard multiple linear regression method. The comparison of the neural network method with the multiple linear regression method, using this data set, revealed that for complex nonlinear problems such as the correlation of the CN with the hydrocarbon type characterization, the neural network approach could provide a better model. However, for a simpler correlation problem like the density of a diesel fuel, which is approximated well by the sum of the contributions of individual components, the predictive equations produced by multiple linear regression and neural network methods gave similar results.     

基于迭代启发式动态规划的锅炉燃烧系统的优化控制

对锅炉燃烧系统进行径向基函数(RBF)神经网络建模,利用基于RBF神经网络的迭代启发式动态规划(HDP)算法对锅炉燃烧系统进行优化控制,并对神经网络初始权值和效用函数进行改进后与传统的HDP算法进行比较分析.