利用结构逼近式混合神经网络实现间歇反应器的建模

提出了一种新的混合神经网络建模方法——结构逼近式混合神经网络。基于此结构建立的混合神经网络可以充分利用已知非线性系统的结构信息,使神经网络“灰盒”化,更好地解释和描述系统各变量间的因果关系,从而提高网络的建模精度和模型的可靠性。本文介绍了这类神经网络的基本特性、拓扑结构和训练方法。报告了一个典型放热液相二级平行间歇反应的建模过程;并针对间歇反应过程测量滞后的情况,与两种不同的混合神经网络模型作了比较,仿真和比较结果证明了方法的有效性。     

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。     

基于混合神经网络的多机驱动系统的协调控制

采用基于PLC的神经网络参数自学习PID控制器,通过神经网络的自学习,对PID参数进行在线实时调整,较好地解决了大型生产输送系统运行过程中的功率不平衡现象。仿真结果表明:混合神经网络PID方法的响应速度快、灵活性好,使电机获得了较好的跟随性能和跟随精度。     

基于计算机方法的混凝土识别与应用研究

为提高混凝土裂缝识别检测的效率,建立卷积神经网络模型,并结合MATLAB对混凝土裂缝形态开展识别。当图像进行灰度处理后,通过理论公式得到了图片特征信息,分析出混凝土裂缝识别效率与二值化阈值的关系,建议二值化阈值取0.3～0.6之间;采用3种检测算子进行裂缝边缘检测,通过对比得到了各检测算子的适用范围:拉普拉斯算子更适用于灰度不同区域,索贝尔算子的识别效率与二值化阈值有关,边缘检测算子能够全面的反映出边缘裂缝情况。卷积神经网络图像识别方法能够为混凝土结构的安全状态评估提供理论和技术支持。     

一种基于神经网络滤波的广义互相关时延估计方法的设计

针对在混合噪声环境中,在未知噪声先验信息的情况下提高基于广义互相关时延估计方法的准确性和适应性的问题,提出了基于神经网络滤波的广义互相关时延估计方法。该方法通过对多个具有特定统计特征的预滤波器进行组合优化来实现对混合噪声的滤波。该时延估计方法具有自调整能力,能够适应动态变化的环境,还能够通过相关函数峰值的大小来优化神经网络滤波器,提高了适应性和精度。     

算子逼近在过程神经网络动态预测中的研究与应用

从泛函分析的角度出发,将动态预测的问题看作算子逼近问题,分析并证明算子逼近能力,进而提出了过程神经网络的动态预测方法,并将时间序列预测问题转化为泛函逼近或算子逼近问题,证明了过程神经网络能以任意精度逼近任意连续算子。最后以Mackey-Glass混沌时间序列预测和打浆度的预测为例,验证该方法的有效性和实用性。     

基于混合评价指标的自组织模糊神经网络设计研究

针对在无增长和修剪阈值时模糊神经网络结构难以自适应问题,提出一种基于混合评价指标(hybrid evaluation index, HEI)的结构设计方法。首先,通过模糊C均值聚类算法(fuzzy C-means clustering, FCM)确定初始规则层神经元数目及其中心与宽度。其次,基于戴维森堡丁指数(Davies bouldin index, DBI)和邓恩指数(Dunn index, DI)提出一种新的相关性评价指标(relevance evaluation index, REI)来计算规则层各神经元输出之间的相关性,同时根据训练过程中网络输出均方根误差(root mean square error, RMSE)的变化情况来确定网络的学习能力,然后基于REI和RMSE提出了HEI。通过HEI来调整模糊神经网络的拓扑结构,有效解决了在无增长和修剪阈值时网络结构难以动态自调整的问题且避免了网络结构冗余。最后,通过对Mackey-Glass时间序列预测、非线性系统辨识和大气中PM_2.5浓度预测,证明了该结构设计方法的可行性和有效性。     

基于混合评价指标的自组织模糊神经网络设计研究

针对在无增长和修剪阈值时模糊神经网络结构难以自适应问题,提出一种基于混合评价指标(hybrid evaluation index, HEI)的结构设计方法。首先,通过模糊C均值聚类算法(fuzzy C-means clustering, FCM)确定初始规则层神经元数目及其中心与宽度。其次,基于戴维森堡丁指数(Davies bouldin index, DBI)和邓恩指数(Dunn index, DI)提出一种新的相关性评价指标(relevance evaluation index, REI)来计算规则层各神经元输出之间的相关性,同时根据训练过程中网络输出均方根误差(root mean square error, RMSE)的变化情况来确定网络的学习能力,然后基于REI和RMSE提出了HEI。通过HEI来调整模糊神经网络的拓扑结构,有效解决了在无增长和修剪阈值时网络结构难以动态自调整的问题且避免了网络结构冗余。最后,通过对Mackey-Glass时间序列预测、非线性系统辨识和大气中PM_(2.5)浓度预测,证明了该结构设计方法的可行性和有效性。     

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。     

基于自组织模糊神经网络的污水处理过程溶解氧控制

针对活性污泥污水处理过程溶解氧浓度控制问题,提出一种基于自组织模糊神经网络（SOFNN）的控制方法。该神经网络控制器依据激活强度和互信息理论在线动态增长和修剪规则层神经元,以满足实际工况的动态变化。同时,采用梯度下降算法在线优化隶属函数层中心、宽度和输出权值,以保证SOFNN的收敛性。进一步通过Lyapunov稳定性理论对SOFNN学习率进行分析,给出控制系统稳定性证明。最后在国际基准仿真平台BSM1上进行实验验证。实验结果显示,与PID、模糊逻辑控制（FLC）和固定结构FNN等控制策略相比,SOFNN在跟踪精度、控制平稳性和自适应能力上更具有优势。     

Control of a Batch Polymerization System Using Hybrid Neural Network ‐ First Principle Model

In this work, the utilization of neural network in hybrid with first principle models for modelling and control of a batch polymerization process was investigated. Following the steps of the methodology, hybrid neural network (HNN) forward models and HNN inverse model of the process were first developed and then the performance of the model in direct inverse control strategy and internal model control (IMC) strategy was investigated. For comparison purposes, the performance of conventional neural network and PID controller in control was compared with the proposed HNN. The results show that HNN is able to control perfectly for both set points tracking and disturbance rejection studies.     

Modeling of thermal conductivity and density of alumina/silica in water hybrid nanocolloid by the application of Artificial Neural Networks

In this research work, the thermal conductivity and density of alumina/silica(Al_2O_3/SiO_2) in water hybrid nanofluids at different temperatures and volume concentrations have been modeled using the artificial neural networks(ANN). The nanocolloid involved in the study was synthesized by the two-step method and characterized by XRD, TEM, SEM–EDX and zeta potential analysis. The properties of the synthesized nanofluid were measured at various volume concentrations(0.05%, 0.1% and 0.2%) and temperatures(20 to 60 °C). Established on the observational data and ANN, the optimum neural structure was suggested for predicting the thermal conductivity and density of the hybrid nanofluid as a function of temperature and solid volume concentrations. The results indicate that a neural network with 2 hidden layers and 10 neurons have the lowest error and a highest fitting coefficient o thermal conductivity, whereas in the case of density, the structure with 1 hidden layer consisting of 4 neurons proved to be the optimal structure.     

Partition coefficient prediction of Baker's yeast invertase in aqueous two phase systems using hybrid group method data handling neural network

A hybrid GMDH neural network model has been developed in order to predict the partition coefficients of invertase from Baker's yeast. ATPS experiments were carried out changing the molar average mass of PEG(1500–6000 Da), p H(4.0–7.0), percentage of PEG(10.0–20.0 w/w), percentage of MgSO_4(8.0–16.0 w/w), percentage of the cell homogenate(10.0–20.0 w/w) and the percentage of MnSO_4(0–5.0 w/w) added as cosolute. The network evaluation was carried out comparing the partition coefficients obtained from the hybrid GMDH neural network with the experimental data using different statistical metrics. The hybrid GMDH neural network model showed better fitting(AARD = 32.752%) as well as good generalization capacity of the partition coefficients of the ATPS than the original GMDH network approach and a BPANN model. Therefore hybrid GMDH neural network model appears as a powerful tool for predicting partition coefficients during downstream processing of biomolecules.     

利用离散正交多项式组合神经网络建立聚合物分子量分布灰箱模型

聚合物分子量分布（MWD）是反映产品性能最重要的指标之一,它是典型的二元建模对象,聚合物分子量分布（MWD）是反映产品性能最重要的指标之一,它是典型的二元建模对象,采用组合神经网络对MWD的空间和时间变量进行分解建模。首先利用离散正交多项式神经网络在链长空间上建立分布与链长的模型,然后将MWD与时间变量的关系转换为网络权向量与输入变量之间的函数,利用递归神经网络建立两者之间的模型,最后组合两个网络达到建模目标。分布函数的模型表达式可写成状态方程形式,为进一步设计控制策略提供了基础。在链长空间上建立模型时,实现了神经网络的权向量与MWD相应阶次矩值之间的等价关系,网络权向量由单纯的拟合数据转变为有意义的物理量,实现了神经网络模型的灰箱化,为精确预测网络隐层节点数问题提供了解决途径。提出的方法应用于实验室规模的苯乙烯聚合过程,证明了建模方法的可行性,同时网络权值与矩值的等价关系也得到验证。     

基于小波网络的非线性动态过程建模和在环氧氯丙烷生产过程中的应用

A type of wavelet neural network, in which the scale function is adopted only,is proposed in this paper for non-linear dynamic process modelling.Its network size is decreased significantly and the weight coefficients can be estimated by a linear algorithm.The wavelet neural network holds some advantages supeiior to other types of neural networks.First, its network structure is easy to specify based on its theoretical analysis and intuition.Secondly, network training does not rely on stochastic gradient type techniques and avoidd the problem of poor convergence or undesirable local minima.The excellent statistic properties of the weight parameter estimations can be proven here.Both theoretical analysis and simulation study show that the identification method is robust and reliable. Furthermore,a hybrid network structure incorporating first-principle knowledge and wavelet network is developed to solve a commonly existing problem in chemical production processes.Applications of the hybrid network to a practical production process demonstrates that model generalisation capability is significantly improved.     

Predicting the vapor-liquid equilibrium of carbon dioxide+alkanol systems by using an artificial neural network

A multi-layer feed-forward artificial neural network has been presented for accurate prediction of the vapor liquid equilibrium (VLE) of CO₂+alkanol mixtures. Different types of alkanols namely, 1-propaol, 2-propanol, 1-butanol, 1-pentanol, 2-pentanol, 1-hexanol and 1-heptanol, are used in this study. The proposed network is trained using the Levenberg-Marquardt back propagation algorithm, and the tan-sigmoid activation function is applied to calculate the output values of the neurons of the hidden layers. According to the network’s training, validation and testing results, a six layer neural network is selected as the best architecture. The presented model is very accurate over wide ranges of experimental pressure and temperatures. Comparison of the suggested neural network model with the most important thermodynamic correlations shows that the proposed neuromorphic model outperforms the other available alternatives. The predicted equilibrium pressure and vapor phase CO₂ mole fraction are in good agreement with experimental data suggesting the accuracy of the proposed neural network model for process design.     

混杂模型模拟甘油发酵过程

以1,3-丙二醇为研究对象,首先建立了甘油发酵的混合半参数模型。混杂半参数模型就是机理与经验模型之间的平衡,特别适用于复杂系统,当然也包括非常复杂的生物系统,可以不通过机理阐明而得到大量数据,近来受到人们的重视。根据建立好的混杂模型,我们利用提前终止法提高了BP神经网络的泛化能力,最终确定了隐含层神经元个数为8,训练函数为traingdx。根据已经建立好了的人工神经网络结构,利用ODE45函数对间歇和批示流加过程混杂模型的微分方程组进行积分求解,最后得到了较好的模拟结果。