时延神经网络的NOx排放动态建模  

Dynamic Modeling of NOx Emission Based on Time-Delay Neural Network

作　　者：张卫庆[1] 柯炎 李凡军[3] ZHANG Weiqing;KE Yan;LI Fanjun(Jiangsu Frontier Electric Technologies Co.,Ltd.,Nanjing 211102,China;Guohua Chenjiagang Electric Power Generation Co.,Ltd.,Yancheng 224005,China;School of Mathematical Science,University of Jinan,Jinan 250022,China)

机构地区：[1]江苏方天电力技术有限公司,江苏南京211102 [2]国华江苏陈家港发电有限公司,江苏盐城224005 [3]济南大学数学科学学院,山东济南250022

出　　处：《自动化仪表》2020年第1期16-21,共6页Process Automation Instrumentation

基　　金：山东省自然科学基金资助项目(ZR2017MF013)

摘　　要：针对复杂工业过程中NOx排放参数建模及预测的问题,引入时延辅助变量,将非稳态工业过程中容易测量失真的辅助变量时间信号转化为空间状态。采用径向基函数(RBF)时延神经网络,利用与预测参数相关的辅助变量和时延辅助变量构建NOx排放的动态模型。仿真结果表明:选取重要辅助变量(总燃料量)的时延单元作为时延辅助变量,通过神经网络训练,可以获得包含非稳态燃烧过程丰富特性的参数模型;设定相同条件,与静态模型进行比较。结果表明,动态模型的内部神经元个数明显少于静态模型,模型结构更紧凑,训练时间更短,泛化能力更强。该方法能提高复杂非稳态工业过程热力参数模型的预测精度,具有较高的应用价值。In order to achieve the modeling and prediction of NOx emission parameters in complicated industrial processes,a time-delay auxiliary variable is introduced to transform the time signals of auxiliary variables in unsteady industrial processes into spatial states.The radial basis function(RBF) time-delay neural network is used to construct dynamic models of important thermodynamic parameters by using auxiliary variables and delay auxiliary variables related to prediction parameters.The simulation results show that the time delay unit of the important auxiliary variable(total fuel volume) can be selected as the time delay auxiliary variable,and the parameter model containing the rich characteristics of the unsteady combustion process can be obtained by training the neural network.Under the same conditions,compared with the static model.The results show that the number of internal neurons in the dynamic model is obviously less than that in the static model,and the structure of the model is more compact,and the training time is shorter.The time is shorter and the generalization ability is stronger.The method can effectively realize the modeling of thermodynamic parameters of complicated unsteady industrial processes,and has great practical value.

关 键 词：燃煤机组 热工参数 时延 神经网络 动态建模 NOX排放 

分 类 号：TH81[机械工程—仪器科学与技术]