A method for building a soft measurement model for dioxin (DXN) emission concentration on the basis of a selective ensemble (SEN) of multi-source potential features. The method comprises: dividing, according to an industrial process, an MSWI process data into sub-systems of different sources, performing principal components analysis (PCA) to respectively extract potential features therefrom, and performing initial selection of multi-source potential features according to experience-based presets of principal component contribution rate thresholds; using mutual information (MI) to measure correlation between the initially selected potential features and DXN, and adaptively determining upper limits, lower limits and thresholds for re-selection of the potential feature; and using a least squares support vector machine (LS-SVM) algorithm having a hyper-parameter adaptive selection mechanism to build DXN emission concentration sub-models for different sub-systems on the basis of re-selected potential features, using a strategy based on a branch and bound (BB) method and a prediction error information entropy weighted algorithm to perform optimal selection of a sub-model and calculate a weight coefficient, and constructing a soft measurement model for DXN emission concentration on the basis of an SEN.
展开▼
