机器学习算法改进及在化工故障诊断中的应用

研究基于机器学习算法,提出一种改进CNN网络的化工故障诊断方法。通过分析CNN网络结构特点与参数训练过程,采用PSO算法对CNN网络进行改进;然后,基于改进CNN网络,提出化工故障诊断方法;最后,通过以TE过程仿真软件,对本研究改进CNN算法在化工故障诊断中的应用进行验证。结果表明：可有效诊断化工故障,平均故障检出率达到91.23%,误报率为1.23%。相较于标准CNN算法、PCA算法、KPCA算法、MICA算法,改进CNN算法对化工故障的检出率更高,误报率更低,且故障检出速度更快。     

基于改进极限学习机的透明塑料裂痕检测

针对透明塑料中微小裂痕难以检测的问题,提出了基于改进的极限学习机算法的检测方法,采用卷积神经网络以组建特征提取器;同时,采用基于狮群算法优化的改进极限学习机算法以构建分类器。在改进极限学习机算法中,狮群算法被用于优化隐含层神经元和输入层神经元之间的权重矩阵,提高了透明塑料微小裂痕检测实验中的识别率。     

基于基团贡献神经网络集成法估算有机物常压凝固点

基于基团贡献法应用人工神经网络对有机物的常压凝固点进行了估算,输入参数为有机物的基团数和表征异构体的参数,输出为常压凝固点.分析了采用最速梯度下降法的BP算法在训练过程中产生误差饱和情况的原因,采用在隐含层节点中加入误差饱和预防函数用来防止误差饱和情况的出现.仿真结果表明,所采用的方法能有效地减小网络在误差表面陷入低谷的可能性和提高网络的收敛速率.采用神经网络集成法建立了神经网络集成模型,通过仿真合理选择隐含层节点数和采用交叉验证法用于防止BP网络的过度训练,增强了网络的泛化能力.估算结果表明,所建立的神经网络集成模型,其网络有良好的稳定性和预测精度,207个样本估算的绝对平均相对误差为8.62%.     

自增长混合神经网络及其在燃料电池建模中的应用

为了提高非线性辨识的精度, 提出了一种基于混合算子的自增长混合神经网络。该神经网络通过自增长的混合隐含层结构, 包括加算子和乘算子, 形成神经元个数少、结果精确、增长快速的网络。论文在级联神经网络的结构基础上, 提出GQPSOI算法来引导神经网络的结构自增长以及权值更新。通过对燃料电池的建模与比较分析, 证明了方法的有效性和良好的应用前景。     

基于BP神经网络的旋风分离器分割粒径模化与预测

为精确建立分割粒径与旋风分离器结构参数和操作参数之间的复杂映射关系,发展了基于数据驱动的BP神经网络（BPNN）的分割粒径模型。使用全局量纲分析,提出环形空间雷诺数、表征旋风分离器本体尺寸影响的量纲为1数和排气芯管插入深度尺寸比作为网络输入参数,表征空气动力等效分割粒径大小的量纲为1尺寸作为网络输出参数,分别确定了训练算法和隐含层神经元个数对BPNN分割粒径模型预测精度的影响。结果表明：贝叶斯正则化算法优于L-M算法和拟牛顿算法,并在隐含层神经元个数为7时达到最优预测性能。与理论模型、半经验模型和多元回归模型进行比较,结果表明,贝叶斯正则化BPNN分割粒径模型展现出了较好的预测能力和泛化性能,模型预测的均方误差为0.136、决定系数为0.975。     

基于特征子空间的KPCA及其在故障检测与诊断中的应用

针对标准KPCA(kernel principal component analysis)不适合大样本分析的缺点,提出了一种基于特征子空间的KPCA(FS_KPCA)及其故障检测与诊断方法,该方法通过构建具有较小维数的特征子空间上的正交基来简化核矩阵,从而降低KPCA的计算复杂性.与标准KPCA方法相比,FS_KPCA方法具有更高的计算效率且只需较小的计算机存储空间.通过非等温连续反应釜过程的故障检测与诊断的应用实例,说明了本算法的有效性.     

前传神经网络规模优化的快速剪枝策略及其应用

对多层前传神经网络规模的优化主要在于确定隐含层的节点数 .本文提出的快速剪枝法从分析网络隐含层的输出信息入手 ,用特征分析法解析地确定冗余而可剪去的隐节点数 ,并一次性地找出优化的隐节点数 ,同时将剪去节点的作用分配到保留的节点上 ,配置结构优化网络的初始连接权 ,以加速网络的训练进程 .这种快速剪枝的解析算法每步都有明确的数学机理 ,不仅优化速度快 ,而且稳定性好 .该算法应用于留兰香油的模式分类问题 ,效果令人满意 ,并显示出良好的健壮性和通用性     

基于KECA的化工过程故障监测新方法

针对化工过程数据复杂、非线性的特点,提出一种基于核熵成分分析(KECA)的化工过程故障监测算法。首先,KECA算法按照Renyi熵值的大小选取特征值及特征向量,相比传统的KPCA监测算法,其保留主元个数更少,可以有效减少运算量。同时,仿真研究表明KECA算法选取的主元具有角度结构特性,据此,提出一种新的统计量--CS(Cauchy-Schwarz)统计量,其对应到核特征空间中即为向量间的角度余弦值,可以较好表述不同概率密度分布之间的相似度。最后,将KECA和KPCA算法分别应用于TE(Tennessee Eastman)过程,结果表明KECA在故障检测延迟与检出率相比KPCA都有很大的优势。     

基于改进聚类算法的RBF网络及其应用

RBF网络可以逼近任意连续非线性函数,且训练速度快,性能好,被广泛应用于过程建模和预测。RBF网络的一个重要因素是隐层节点的选择,隐层节点过多或过少都会影响最终网络的性能。提出一种改进的k-means聚类算法,可以自动确定最优的聚类区数,并且可使最终的聚类中心合理地分布在数据空间中。在应用RBF网络进行建模和预测时,采用该方法确定隐层节点的中心,跟用通常的聚类方法相比,可以大大减小网络规模。仿真和实际应用结果都证明该方法的有效性。     

基于自动谱聚类与多极端学习机模型的油井油液含水率软测量

油井油液的含水率是石油生产中的一个重要参数,及时、准确的测量对提高采油生产效率具有重要的意义。针对传统人工测量所存在的不足,引入软测量技术,建立基于自动谱聚类与多极端学习机（automatic spectral clustering-multiple extreme learning machines, ASC-MELM）的软测量模型。提出一种自动谱聚类（spectral clustering, SC）算法,由改进的萤火虫算法（firefly algorithm, FA）对聚类数目和尺度参数进行优化选取,所提出的改进萤火虫算法（improved firefly algorithm, IFA）采用以一定概率跳出当前解的机制,避免传统FA过早陷入局部最优解的不足;对聚类后的不同训练子集,分别由极端学习机（extreme learning machine, ELM）建立子模型,由IFA对其中的隐含层输入权值、隐含层神经元的偏置和隐含层节点个数进行优化选取;最后,将多个子模型的结果取加权平均值输出。由国内某油田作业区一口生产井进行实例验证,结果表明所提出方法具有较高的预测精度,对于实现油井油液含水率的软测量是合理有效的。     

基于模糊RBF神经网络的乙烯装置生产能力预测

针对传统的径向基函数(RBF)神经网络隐藏层节点的不确定和初始中心敏感性、收敛速度过慢等问题,提出一种基于模糊C均值的RBF神经网络(FCM-RBF)模型,通过模糊C均值聚类(FCM)得到各聚类中心,基于误差反传的梯度下降法训练隐藏层到输出层之间的权值,克服传统RBF模型对数据中心的敏感性,优化确定RBF神经网络隐藏层的节点数,提高网络训练速度和精度。最后将其用于乙烯装置生产能力预测中,分析预测不同技术、不同规模乙烯装置生产情况,指导乙烯生产,提高生产效率,结果验证了所提出算法的有效性和实用性。     

基于RBF神经网络的油藏相对渗透率曲线计算

提出了一种基于改进的RBF神经网络的相对渗透率曲线计算方法。利用骨干粒子群的位置更新操作更新RNA遗传算法的变异算子得到混合RNA遗传算法（HRGA）,针对RBF神经网络中隐含层径向基中心值的确定,利用HRGA算法对其进行优化,并用于相对渗透率曲线的计算。将HRGA优化的RBF神经网络和标准RBF神经网络计算的相对渗透率曲线与真实值误差对比分析,实验结果表明HRGA优化的RBF神经网络明显提高了计算精度。     

基于剪接系统的遗传算法RBF网络建模方法

A splicing system based genetic algorithm is proposed to optimize dynamical radial basis function （RBF） neural network, which is used to extract valuable process information from input output data. The novel RBF network training technique includes the network structure into the set of function centers by compromising between the conflicting requirements of reducing prediction error and simultaneously decreasing model complexity. The effectiveness of the proposed method is illustrated through the development of dynamic models as a benchmark discrete example and a continuous stirred tank reactor by comparing with several different RBF network training methods.     

A new algorithm for developing dynamic radial basis function neural network models based on genetic algorithms

A new method for extracting valuable process information from input–output data is presented in this paper. The proposed methodology produces dynamical radial basis function (RBF) neural network models based on a specially designed genetic algorithm (GA), which is used to auto-configure the structure of the networks and obtain the model parameters. The new RBF network training technique formulates a complete optimization problem, which includes the network structure into the set of free variables that are used to minimize the prediction error. This is a different approach compared with the local search methods employed by other structure selection mechanisms, which are often trapped to local minima. Another advantage of the proposed method is that only one run of the algorithm is required to obtain the optimal network structure, in contrast to the standard RBF training techniques, where the produced model is selected by trial and error. The effectiveness of the method is illustrated through the development of dynamical models for two sets of data: simulated data from a Continuous Stirred Tank Reactor (CSTR) and true data collected from a Kamyr digester, which is a rather complicated reactor used in the pulp and paper industry.     

Prediction of the gas solubility in polymers by a radial basis function neural network based on chaotic self‐adaptive particle swarm optimization and a clustering method

A novel model based on a radial basis function neural network (RBF NN), chaos theory, self‐adaptive particle swarm optimization (PSO), and a clustering method is proposed to predict the gas solubility in polymers; this model is hereafter called CSPSO‐C RBF NN. To develop the CSPSO‐C RBF NN, the conventional PSO was modified with chaos theory and a self‐adaptive inertia weight factor to overcome its premature convergence problem. The classical k‐means clustering method was used to tune the hidden centers and radial basis function spreads, and the modified PSO algorithm was used to optimize the RBF NN connection weights. Then, the CSPSO‐C RBF NN was used to investigate the solubility of N₂ in polystyrene (PS) and CO₂ in PS, polypropylene, poly(butylene succinate), and poly(butylene succinate‐co‐adipate). The results obtained in this study indicate that the CSPSO‐C RBF NN was an effective method for predicting the gas solubility in polymers. In addition, compared with conventional RBF NN and PSO neural network, the CSPSO‐C RBF NN showed better performance. The values of the average relative deviation, squared correlation coefficient, and standard deviation were 0.1282, 0.9970, and 0.0115, respectively. The statistical data demonstrated that the CSPSO‐C RBF NN had excellent prediction capabilities with a high accuracy and a good correlation between the predicted values and the experimental data. © 2013 Wiley Periodicals, Inc. J. Appl. Polym. Sci. 130: 3825–3832, 2013     

Melt index prediction by RBF neural network with an ICO‐VSA hybrid optimization algorithm

Melt index (MI) is considered as one of the most significant parameter to determine the quality and the grade of the practical polypropylene polymerization products. A novel ICO‐VSA‐RNN (RBF neural network with ICO‐VSA algorithm) MI prediction model is proposed based on radial basis function (RBF) neural network and improved chaos optimization (ICO), and variable‐scale analysis (VSA), where the ICO is first added and then combined with the VSA to overcome the defects of ICO and VSA, then the parameters of the RBF neural network are optimized with them. At last, the RBF neural network model for MI prediction model is developed. Further researches on the optimal RBF neural network model of MI prediction are carried out with the data from a real industrial plant, and the prediction results show that the performance of this prediction model is much better than the RBF neural network model without optimization. © 2012 Wiley Periodicals, Inc. J Appl Polym Sci, 2012     

Intelligent Prediction of CO2 Capture in Propyl Amine Methyl Imidazole Alanine Ionic Liquid: An Artificial Neural Network Model

The objective of this paper is to create a new artificial neural network (ANN) model to predict solubility of CO₂ in a new structure of task specific ionic liquids called propyl amine methyl imidazole alanine [pamim][Ala]. Equilibrium data of CO₂ solubility were measured at the temperatures of 25, 40, and 60°C and the pressures up to 50 bar. For the purpose of performance comparison, the two most common types of ANNs, multilayer perceptron (MLP) network and radial basis function (RBF) network were used. Water content, ionic liquid content, temperature, and pressure set as input variables to ANN, while CO₂ capture rate assigned as output. Based upon optimization process, MLP neural network with 14 neurons in the hidden layer, log-sigmoid transfer function in the hidden layer and linear transfer function in the output layer, exhibited much better performance in prediction task than RBF neural network with the same neuron numbers in the hidden layer. Results obtained demonstrated that there is a very little difference between the estimated results of ANN approach and experimental data of CO₂ capture rate for the training, validation, and test data sets. Furthermore, Henry’s law constants were obtained by fitting the equilibrium data.     

Machine learning modelling and evaluation of jet fires from natural gas processing,storage, and transport

Jet fires and their repercussions play a significant role in catastrophic incidents that typically have a cascading impact in process industries. Several hydrocarbon experiments from 19 papers were incorporated into the current endeavour to develop simulations of jet flames using machine learning (ML) models. Dimensionless characteristics have been used as output and input variables, including mass flow rates, fuel density, jet flame length, and heat release fluxes. When training three layers of the multi-layer feedforward neural network (MLFFNN) method, a Bayesian regularization backpropagation approach was adopted and evaluated with the radial based functions (RBF) algorithm. Through an optimization procedure, the first and second hidden layers of the MLFFNN have been optimized to include 10 and five neurons, respectively. The RBF algorithm with 40 neurons in a single layer has been computed using the same method. The best mean square error (MSE) validation results for RBF and MLFFNN were 0.006 and 0.0002, respectively, for 40 and 100 epochs. The MLFFNN and RBF models' respective regression statistical analysis outputs were 0.9949 and 0.9645. The ML method has been identified as a potentially useful technique for precisely predicting the geometrical and radiative characteristics of jet flames.