Purposes Predicting the ash content of clean coal on the basis of flotation foam images is one of the core methods for intelligent flotation control. To address the limitations of existing methods caused by incomplete feature representation dimensions and insufficient multi-source information fusion, a prediction method that integrates a Multi-channel Attention Regression Network (MARNet) with feature engineering is proposed. Methods First, the MARNet model with multi-color-space image inputs was constructed, leveraging attention mechanisms to foam on distinct foam information from different color spaces. Simultaneously, a feature engineering-based ash prediction model was developed by using the XGBoost algorithm, incorporating morphological, color, texture, and frequency-domain features. A Bayesian dynamic weighted fusion strategy was established to adaptively allocate prediction weights between MARNet and XGBoost on the basis of posterior probability optimization criteria, forming a synergistic prediction mechanism that combines their complementary strengths. Results The experimental results demonstrate that the prediction performance of the MARNet and feature engineering integrated model significantly outperforms traditional methods. The proposed model achieves a coefficient of determination (R2) of 0.969 1, a mean absolute error (MAE) of 0.126 4%, and a root mean square error (RMSE) of 0.216 1%. Compared with XGBoost and MARNet single-model predictions, the RMSE is reduced by 70.90% and 25.22%, respectively, confirming a substantial improvement in prediction accuracy, which validates the effectiveness of the proposed method for intelligent control in flotation processes.
SHANGM.Application of online coal quality detection equipment in intelligent construction of coal preparation plant[J].Clean Coal Technology,2023,29(S1):154-158.
WANGJ, WANGR F, WEIK,et al.Multi-step prediction of heavy medium clean coal ash content based on time series alignment and TCNformer[J].Industry and Mine Automation,2024,50(5):60-66.
FANF B, WANGR F, FUX.Research and application of online ash detection technology for flotation clean coal based on multi-source heterogeneous information fusion[J].Coal Technology,2023,42(2):233-238.
[13]
QIUZ Y, DOUD Y, ZHOUD Y,et al.On-line prediction of clean coal ash content based on image analysis[J].Measurement,2021,173:108663.
BUX Z, YANGY L, WANH.Research progress in predicting concentrate grade based on flotation froth images[J].Metal Mine,2024(2):25-38.
[18]
ZHANGZ L, LIUY, HUQ,et al.Multi-information online detection of coal quality based on machine vision[J].Powder Technology,2020,374:250-262.
[19]
CUIY, ZHANGK H, LVZ Q,et al.Exploring the effect of various factors for ash content estimation via ensemble learning:Color-texture features,particle size,and magnification[J].Minerals Engineering,2023,201:108212.
[20]
WENZ P, ZHOUC C, PANJ H,et al.Froth image feature engineering-based prediction method for concentrate ash content of coal flotation[J].Minerals Engineering,2021,170:107023.
[21]
ZHANGK H, WANGW D, LVZ Q,et al.A CNN-based regression framework for estimating coal ash content on microscopic images[J].Measurement,2022,189:110589.
[22]
LUF, LIUH, LVW.Deep correlation and precise prediction between static features of froth images and clean coal ash content in coal flotation:An investigation based on deep learning and maximum likelihood estimation[J].Measurement,2024,224:113843.
WANH, ZHOUK Q, ZHAOL F,et al.Research on coking coal flotation model based on BP neural network[J].Industrial Minerals & Processing,2023,52(12):8-13.
[27]
WENZ P, ZHOUC C, PANJ H,et al.Deep learning-based ash content prediction of coal flotation concentrate using convolutional neural network[J].Minerals Engineering,2021,174:107251.
[28]
YANGX L, ZHANGK F, NIC,et al.Ash determination of coal flotation concentrate by analyzing froth image using a novel hybrid model based on deep learning algorithms and attention mechanism[J].Energy,2022,260:125027.
HANY, WANGL H, LIUQ S,et al.Intelligent detection method for flotation tailings ash content based on CNN-BP[J].Industry and Mine Automation,2023,49(3):100-106.
[31]
GB/T 30046.1-2013 煤粉(泥)浮选试验 第1部分:试验过程 [S].
[32]
TANM, LEQ.Efficientnet:Rethinking model scaling for convolutional neural networks[C]//International conference on machine learning,2019:6105-6114.
[33]
ZHANGJ, TANGZ H, AIM X,et al.Nonlinear modeling of the relationship between reagent dosage and flotation froth surface image by hammerstein-wiener model[J].Minerals Engineering,2018,120:19-28.
[34]
HARALICKR M.Statistical and structural approaches to texture[J].Proceedings of the IEEE,1979,67(5):786-804.
[35]
ARUNACHALAPERUMALC, DHILIPKUMARS.An efficient image quality enhancement using wavelet transform[J].Materials Today:Proceedings,2020,24:2004-2010.