Coal flotation tailings ash content prediction method based on multi-source feature fusion and FAMFormer
WEN Zhiping
ZHOU Maiqiang
LIU Cheng
ZHOU Changchun
Abstract：The current coal flotation production exhibits extensive operation patterns,with one critical obstacle to precise regulation of flotation systems being the online prediction of tailings ash content.Existing tailings ash content modeling techniques primarily rely on visual model-ing,facing challenges including missing feature information and poor model generalization per-formance,leading to unsatisfactory practical application effectiveness.To address these is-sues,this study proposes a deep network architecture featuring arithmetic feature interaction(FAMFormer)based on a visual feature fusion tabular data.Specifically,a ResNet-SENet deep convolutional network is developed to extract convolutional features from flotation tail-ings images,and a traditional feature engineering is established from four perspectives:gray histogram,co-occurrence matrix,statistical analysis,and color space.A fused tailings ash content dataset is constructed by integrating deep convolutional features with traditional fea-tures.Correlation analysis reveals that deep convolutional features generally exhibit stronger correlations with tailings ash content compared to traditional feature-engineered characteris-tics.The proposed regression model achieves a mean absolute error(EMA)of 0.75 through di-rect training on the fused visual feature dataset.After principal component analysis(PCA),comparative experiments with three classical models demonstrate that the FAMFormer a-chieves optimal prediction performance for coal slime flotation tailings ash content,with EMA of 0.51,ERMS of 0.72,and R2 of 0.98.Finally,an ash content prediction system software is developed and validated through industrial applications.The validation results confirm that the trained model exhibits strong generalization capabilities,providing methodological and theoret-ical support for tailings ash content modeling research in coal slime flotation processes.
Keywords：coal flotationtailings ash contenttailings imageFAMFormerMulti-source fea-ture fusion
Publication Date：2025-07-30
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：13( 930-942 )