Dioxin Emission Concentration Prediction Using the Selective Ensemble Algorithm Based on Bayesian Inference and Binary Tree

Dioxin (DXN), a by-product from the municipal solid waste incineration (MSWI) process, is an organic pollutant; thus, it is extremely harmful to the ecological environment and difficult to detect in real time. A selective ensemble (SEN) model for DXN emission concentration based on Bayesian inference and binary trees is proposed given the weak interpretability, high model complexity, and poor generalization performance of the existing DXN emission concentration prediction model. Initially, bagging sampling is used to obtain different data subsets. The binary tree, as the candidate submodel, is constructed based on the sub-datasets, and the prior information of the leaf nodes and the predicted values of the candidate submodel are calculated. Bayesian inference is used to calculate the posterior information to characterize the fitness of the candidate submodel. Based on the posterior error, the best submodel is selected as the ensemble submodel. These processes are repeated to obtain all the ensembled submodels and the corresponding posterior information. Then, the combined weights are determined by the posterior information of all ensemble submodels, and the DXN emission concentration SEN model is constructed. The effectiveness of the proposed method is verified using the actual data of the MSWI process.