Selective catalytic reduction(SCR)is the most widely used flue gas denitrification technology in coal-fired power plants.The rapid and accurate measurement of NO_x production at the inlet of SCR reactor affects the regulation of ammonia injection and NO_x emission at the outlet of SCR reactor.Because SCR denitration system has the characteristics of non-linear and large delay,it takes a certain time to measure the NO_x generation at the inlet by flue gas analyzer,which leads to the lag between the measured value and the real value,so it is difficult to realize the precise control of SCR denitration system,so it n eeds to get the timely and accurate NO_x generation at the inlet of SCR reactor.Therefore,this paper establishes a data-driven model of NO_x production at the inlet of SCR rea ctor to predict NO_x production.Boiler combustion is a complex industrial process,so it is difficult to build a model by mechanism,it is necessary to build a data-driven model.There are many auxiliary variables that affect the NO_x production at the inlet of SCR reactor and they have certain correlation.There is a certain time lag between the auxiliary variables and NO_x production.To establish an accurate model,it is necessary to accurately select the auxiliary variables and determine their time lag.C ompared with the previous studies,which ignored the time-delay estimation of auxiliary variables or divided the auxiliary variable selection and time-delay estimation into two independent processes,this paper proposes a variable time-delay joint selection method based on joint mutual information,which uses conditional mutual information for variable selection and time-delay estimation,in principle,avoids the influence of the correlation between auxiliary variables on the results and maximizes the auxiliary variable set Contains valid information and minimizes redundant information.Firstly,the factors affecting NO_x production are analyzed,and 30 initial variables are selected.Based on the proposed variable delay joint selection method,the auxiliary variables are selected at the same time the time delay is estimated.In order to verify the effectiveness of the method proposed in this paper,mutual information and conditional mutual information(without delay estimation)are used to select auxiliary variables,and the selection results are used as the input variables of the model to establish a BP neural net work model for comparative study.The results show that the model based on the variable delay joint selection method proposed in this paper has higher accuracy and stronger generalization ability,which proves the superiority of this method.Finally,particle swarm optimization is used to optimize the radial basis parameters and error penalty factors of LSSVM.The auxiliary variable set selected by th e method is used as the input variable,and the NO_x production is used as the output variable to establish t he PSO-LSSVM model.The results show that the model established in this paper is more accurate in NO_x production prediction and has stronger general ization ability. |