A Dual Hybrid Data-Driven Strategy for Soft Measurement of Dioxins Emission

Accurately detecting dioxin (DXN) emission concentration from municipal solid waste incineration (MSWI) process is essential, yet it remains a significant challenge. This article presents a novel dual hybrid data-driven strategy for developing a soft measurement model. The strategy achieves both the virtual-real data hybrid and improved main-compensation structure hybrid, significantly enhancing the interpretability and applicability of soft measurement. A virtual simulation mechanism dataset is generated through an orthogonal experimental design and implementation, based on a coupled numerical simulation model for DXN formation and the operating conditions of an actual MSWI power plant. An enhanced main-compensation structure soft measurement model is developed by fusing interpretable quasi-Newton decision trees (QNDT), interval type-3 fuzzy broad learning systems (IT3FBLS), and variable step size least mean square (VSS-LMS) ridge regression, using both real historical process data and virtual simulation mechanism datasets. The offline model training and online model verification stages are described in detail. Experimental results demonstrate that the strategy significantly outperforms conventional single hybrid data-driven approaches and state-of-art methods in terms of generalization and interpretability.