Research On Prediction Method Of The NO_x Emissions From Power Plants Boiler

In recent years,nitrogen oxide(NOx)emissions of the power plant boiler caused serious environmental pollution.In order to meet environmental protection requirements,and satisfy the demand of power generation enterprises energy conservation and emissions reduction,establish accurate and effective model of NOx emission is the foundation of effective inhibition of NOx emissions.In this paper,in the foundation of the optimized single least square support vector machine(LS-SVM)modeling of NOx emission,a multi LS-SVM-based ensemble model is proposed to predict NOx emissions to obtain good performance.The main work of this paper is the following aspects.(1)In the first part,considering the methods of NOx reduction,illustrates the necessity of accurate forecasts on NOx emissions,the current research of both forecasting methods of NOx emissions and ensemble modeling method.(2)In the second part,introduces the single LS-SVM model of NOx emissions.First of all,the characteristics the test boiler has been elaborated;Second,this paper introduces the test data of bolier and feature variable selection method;Then,this paper introduces the basic principle of LS-SVM;Finally,the single LS-SVM model of NOx emissions is established,and optimization of the super parameters of LS-SVM model uses the genetic algorithm(GA).Compared with LS-SVM model using the traditional grid search method,support vector machine(SVM)model and BP neural network model,the GA-LSSVM model was validated with more accurate model accuracy and better generalization ability.(3)Because of the limitation of the single model,such as low training efficiency and poor generalization ability,the modeling method of multi model integration is discussed in this paper.A supervised clustering algorithm based on improved soft fuzzy C means algorithm which is optimized by genetic algorithm(GA-Soft Fuzzy C-Means,GA-SFCM),the new algorithm improves the stability of the sub models,achieves clear physical meanings of each sub model,and avoids the blindness;at the same time,this paper propose a new clustering feature selection method for regression model,which improves the efficiency of clustering.(4)According to the NOx emissions from low to high,the data space is divided into low,medium and high subspace,multi LS-SVM ensemble model based on GA-SFCM clustering algorithm is established for the NOx emissions,the ensemble model uses least squares method integrates sub models.Compared with models with different clustering methods and integration methods,the simulation results show that the multi LS-SVM ensemble model based on supervised GA-SFCM clustering method improves stability of the clustering results,compared to the single LS-SVM model,ensemble model prediction accuracy and generalization ability are improved.