Bagging based ensemble learning approaches for modeling the emission of PCDD/Fs from municipal solid waste incinerators

Abstract

The conventional method for determining polychlorinated dibenzo-p-dioxins and polychlorinated dibenzofurans (PCDD/Fs) emission concentrations from municipal solid waste incinerators (MSWIs) is accurate but complex, costly, and time-consuming. In this study, we utilized machine learning approaches to model International Toxic Equivalent Quality (I-TEQ) values of PCDD/Fs through chlorobenzene (CBz). Sampled data from a MSWI with a grate furnace and three MSWIs with fluidized bed furnaces were randomly divided into the training and test set. A two-step data preprocessing consists of Box-Cox transformation and standardization was applied to improve the model performances. The models including artificial neural network (ANN) and decision tree (DT). With the help of bagging, ANN and DT were stacked into bagging-based ensemble neural networks (BGNN) and random forest (RF) to improve the accuracy and robustness. BGNN was proved to be the best model with the relatively good perfomance in the test set. The Shapley additive explanation (SHAP) method was employed to improve the interpretability of BGNN. The results show that 1,2,3-trichlorobenzene (TrCBz), 1,4-dichlorobenzene (DiCBz), 1,2,4-TrCBz, and 1,3-DiCBz make positive contributions to I-TEQ values. Employing machine learning models, particularly BGNN, can be an effective method to determine the emission of PCDD/Fs within a wide range of I-TEQ values and in multiple MSWIs with strong generalization ability. It is helpful to realize the on-line measurement and controlling of PCDD/Fs emission.

Keywords: Chlorobenzene (CBz); Data preprocessing; Machine learning; Model interpretation; Polychlorinated dibenzo-p-dioxins and polychlorinated dibenzofurans (PCDD/Fs).