Evaluation of parametric effects of coal flotation based on boosting modeling method

Accurate assessment of the effects of parameters on the flotation process is important for understanding the complex flotation mechanisms. To address the problem of unsatisfactory prediction of large sample flotation data (641 sets) by traditional machine learning algorithms, four advanced algorithms (GBDT, CatBoost, LightBGM and XGBoost) are used in this paper to investigate the effects of feed properties and flotation conditions on the effectiveness of coal flotation. It was found that the data at flotation recoveries below <40% were difficult to predict effectively by machine learning algorithms due to abnormal flotation results caused by lower flotation reagent dosages. An importance analysis of flotation parameters and prediction of flotation results were carried out based on the reordered data. The results showed that the fraction and ash content of -74 um in the feed are the main factors affecting concentrate yield and ash content. The XGBoost model also achieved the best prediction results compared to other models, and the prediction coefficient of determination R2 reached 0.877 and 0.971 for concentrate yield and ash content, respectively. The results are expected to provide a reference for the intelligent control of coal beneficiation plant by machine learning technology in the future.