Prediction of Dioxin Emission Concentration in the Municipal Solid Waste Incineration Process Based on Optimal Selection of Virtual Samples

Municipal solid waste incineration (MSWI) process emits dioxins (DXN), which are known as "poison of the century". The long period and high cost off-line detection method is adopted in the industrial field. Thus, the samples for constructing the DXN emission concentration prediction model are scarce. To solve above problems, this paper proposed a modeling strategy and its corresponding modeling method based on virtual sample optimization selection. First, on the basis of pre-processing such as outlier removal and sample input/output matching, the reduced small sample data was obtained by feature selection combined with process characteristics and mechanism knowledge. Then, the input/output domain of the reduced small sample was extended based on expert knowledge and the mega-trend-diffusion technique. Then, the virtual sample inputs generated based on the mechanism knowledge and interpolation algorithm, and the virtual sample outputs were obtained based on the mapping model constructed by the reduced small sample. Further, the virtual samples were reduced in combination with the extended input/output domain to obtain the candidate ones. Then, based on particle swarm optimization (PSO) algorithm, the candidate virtual samples were optimally selected. Finally, a prediction model was constructed by using a mixture of selected virtual samples and reduced training samples. Combined with the DXN data of an incinerator, the validity of the proposed method was verified.