Research On Prediction Modeling Method For NO_x Emissions From Boiler Flue Gas Of Power Plant

The emissions of nitrogen oxides NO_xfrom power plant boilers is of great significance to the economic benefits and environmental management of power plants.Therefore,accurate and reliable prediction of smoke NO_xemissions is essential for low-nitrogen operation of boilers.However,traditional prediction models with low prediction accuracy and low generalization ability cannot be adapted to the complex nonlinear system.Based on the single prediction model and multi-model ensemble prediction model,the paper improves prediction method which is the Least Squares Support Vector Machine(LSSVM),and designs three prediction algorithms.The data roots in the#2 unit of a power plant is verified to improve the prediction accuracy of NO_xemissions.The main research work of the thesis are as follows:(1)The current situation that the NO_xemissions from boiler flue gas is predicted and modeled by using single model and ensemble model at home and abroad is studied.And then the characteristics of NO_xemissions for boilers are analyzed.(2)A single model prediction modeling method based on Whale Optimization Algorithm-Least Squares Support Vector Machine(WOA-LSSVM)is proposed.Firstly,the initial sample data is normalized,and then the two parameters of the kernel function widthσ~2and the penalty factor c in LSSVM are optimized by the Whale Optimization Algorithm(WOA)algorithm to establish WOA-LSSVM model.And finally get the model output.(3)A multi-model ensemble prediction modeling method based on Modified Whale Optimization Algorithm-Least Squares Support Vector Machine(MWOA-LSSVM)is proposed.Firstly,the sample space is divided,and then the segmentation Logistic chaotic map is used to initialize the population,the control variable a is improved,nonlinear adaptive parameters is introduced and Quadratic Interpolation(QI)used to update position,etc.to improve the global exploration ability of WOA.The MWOA is used to globally optimizeσ~2and c of the LSSVM sub-model on each subspace to obtain the sub-model output.Finally,the output of each LSSVM sub-model is integrated as the output of the ensemble model.(4)A multi-model ensemble prediction modeling method based on Multi-model Fuzzy Cluster Ensemble algorithm(MFCE-LSSVM)is proposed.Firstly,the data space is divided according to the level of output NO_xemissions.The variables participating in the clustering are determined by the weight of the correlation analysis and the hierarchical clustering based on information entropy,and then data is clustered by the proposed Multi-model Fuzzy Cluster Ensemble(MFCE)algorithm to obtain the membership matrix of each subspace.And the LSSVM model of each subspace is integrated by the least squares method fused membership degree,and finally the output of the ensemble model is obtained.Finally,the three proposed methods is used to simulate the selected experimental boiler in order to verify the effectiveness of the methods.At the same time,other prediction models are selected as comparative studies in order to verify the superiority of the proposed methods.The simulation and comparison results show that all three methods can effectively predict the NO_xemissions of boilers,and the integrated MWOA-LSSVM prediction method is complete with the most stable and the most accurate prediction performance.