Research On The NO_x Soft Sensor Model Of SCR Flue Gas Denitrification System Based On Data Driven

The power industry is a pillar industry of the national economy,and thermal power generation occupies a dominant position.It is of great environmental protection significance to limit the emission of air pollutants from thermal power units.The optimal control of the SCR flue gas denitrification system and the accurate measurement of the inlet Nitrogen Oxides(NO_x)concentration are the current challenges in the thermal power field.Based on the data-driven soft sensing technology and the actual data of thermal process,the following researches are carried out respectively:(1)Aiming at the characteristics of high dimension and nonlinearity of measured variables in the actual thermal process,and the subjectivity of existing variable selection methods,the application of the Elastic Net method in the soft sensing model of SCR inlet NO_x concentration was explored.Combined with the advantages of the least squares support vector regression(LSSVR)model,the Enet-LSSVR model is proposed.The experimental results show that the proposed method effectively reduces the dimension of variables,and the soft sensing model has a high level of accuracy and realizes the prediction of NO_x concentration at the inlet of the SCR flue gas denitrification system.(2)According to the application characteristics of the actual engineering process,combined with the advantages of the Elastic Net model,the Elastic Net variable selection model integrating the engineering mechanism is further proposed.The experimental results show that,compared with the measurement variables of the Elastic Net method,the selected feature variables of the fusion engineering mechanism have lower dimension and higher prediction accuracy,and have better engineering application value.(3)Aiming at the failure of the static soft-sensor model in the actual thermal process,a dynamic soft-sensor modeling method under variable working conditions is studied.The time-delay estimation of the feature variables is realized by the complex correlation coefficient time-delay joint method,and the modeling sample is updated by the sliding window update strategy(Moving Window,MW).In order to combine the analysis advantages of different soft-sensor models,the prediction strategy of the variable weight combination model is further explored.The experimental results show that,compared with the static model,the proposed model update strategy can effectively improve the prediction accuracy of NO_x concentration,the obtained soft sensor model has high stability and generalization,and has excellent analysis results in the process of actual variable operating conditions.