Research on soft sensing of dioxin emission concentration from garbage incinerators based on machine learning
The concentration of dioxin in the flue gas of waste incinerators has been the focus of attention.Measurement of dioxin has the problems of high cost,long time,and inability to monitor in real time,so it is particularly important to realize the indirect online monitoring of dioxin concentration through soft measurements based on the operating parameters of the incinerator and the concentration of the pollutants that can be monitored.In this paper,six different types of regression prediction models,namely,support vector machine,extreme Gradient Boosting(eXtreme Gradient Boosting(XGBoost),random forest,neural network,decision tree,and linear regression,are established using machine learning algorithms to improve the generalization ability by optimizing the parameters and decreasing the model error,and five folds of the Cross-validation method and regression model evaluation criteria were used to comprehensively analyze and compare these six models.The results showed that the support vector machine model performed the best for both the training and test sets,and the mean absolute error(SMAE),relative error(SRE),and coefficient of determination(R2)for the test set were 6.93,3.36,and 0.98,respectively.Five characteristic variables,namely,the flue gas temperature,the combustion chamber temperature,the CO concentration,the HCl concentration,and the particulate concentration,were analyzed by the support vector machine model for their effects on dioxin concentration.The effects on dioxin concentration were analyzed by support vector machine modeling,and it was found that CO concentration had the greatest degree of influence on dioxin concentration,which was positively correlated;combustion chamber temperature was the next highest,and the dioxin concentration was the greatest when the combustion chamber temperature was in the range of 800～900℃.This study provides a theoretical basis for the soft measurement of dioxin emission concentration in waste incinerators.
万方数据
维普
