Mutual Information Based Research On Dynamic Modeling Of SCR System

Coal-fired power generation dominates in the power industry of China,and nitrogen oxides(NO_x)produced by coal-fired power plants have caused serious pollution on environment.Selective Catalytic Reduction(SCR)is an effective method to remove NO_x from the flue gas of power plants,and the establishment of an accurate SCR denitrification system model is the basis of SCR denitration.In recent years,the development of informatization of power plants makes it easier to obtain process data.However,unprocessed high-dimensional data contains unnecessary information and complex relationships,which will cause dimensionality curse and increase the complexity of modeling.According to the SCR denitrification system,this article use mutual information to select the correlative variables which affect the NO_x concentrations at the outlet of the SCR reactor,and the selected variables are used as inputs in the modeling of SCR denitrification system.The results show that the model has achieved a good performance in following and predicting the dynamic changes of SCR denitrification system.The main work of the paper is as follows:(1)Firstly,the background of denitrification system and its meaning are introduced.Secondly,the research methods,status of variable selection and system modeling are discussed,and the structure,process flow and reaction principles of SCR denitrification system are investigated in detail.(2)The SCR denitrification system is a complex nonlinear system.There are many factors affecting NO_x at the outlet of SCR reactor,and exists some interferences from redundant variables.On the basis of mutual information theory,this paper proposes a method of variable selection based on stepwise-weighted mutual information.Firstly,the time series of every input variable are analyzed by mutual information to ensure the pure delay time.Then a dynamic weight coefficient is used to improve the accuracy of the selection of input variables.(3)Taking the model selected by mutual information as input variables,we construct the model of SCR denitrification system by extreme learning machine,and optimize the model parameters by genetic algorithm.The results show that the algorithm has a certain improvement in forecast accuracy and generalization ability.