Hybrid Simulator-Based Mechanism and Data-Driven for Multidemand Dioxin Emissions Intelligent Prediction in the MSWI Process
Heng Xia, Jian Tang, Wen Yu, Junfei Qiao
Abstract
The real-time detection technique and comprehensive characterization of dioxin (DXN) emission concentration during the municipal solid waste incineration process persist as unresolved challenges. Prevailing research predominantly relies on data-driven models, often overlooking the potential benefits derived from fusing combustion mechanism knowledge. To confront this issue, we propose a hybrid modeling strategy that fuses a simulator-based mechanism model with an enhanced regression decision tree-based data model. This approach aims to predict DXN emission concentrations while accommodating diverse time-scaled measurement requirements. Based on virtual mechanism data obtained via numerical simulation models coupling FLIC and Aspen Plus, we constructed a white-box surrogate model utilizing a multiple-input multiple-output linear regression decision tree (LRDT). To establish a relationship with DXN emission concentration, we employed a semisupervised transfer learning mapping model. It was then fused with a novel ensemble LRDT model based on real historical data by using a constrained incremental random weight neural network. The efficacy of this modeling strategy was validated through an industrial application case study conducted in Beijing.