Abstract
The intelligent design of copper flotation processes is an important means for improving resource utilization and reducing costs in the current mining industry. In our previous study, the flotation process design is split into the backbone process design and the cleaning and scavenging process design. The copper flotation backbone process design was transformed into multi-label classification, and it was found that applying label correlation and domain knowledge to multi-label classification could significantly improve the precision of backbone process design. However, the above method cannot be used for cleaning and scavenging process design, which is a multi-label regression problem. This study improves on our previous work and proposes new methods that allow it to simultaneously handle multi-label classification and regression for the design of copper flotation processes. Moreover, further improves the prediction effect through the following methods: (1) Referencing adaboost algorithm, the training set samples with large prediction error in the previous iteration are set with higher weight; (2) To enhance robustness, the label uncertainty coefficient is introduced; (3) To alleviate the over fitting of small sample machine learning, bootstrap aggregating is introduced for each sub label. The experimental results demonstrate the significantly superiority of the proposed method in the copper flotation process design. The above research has significant implications for the engineering application of artificial intelligence. From the ablation experimental results, these improvements are effective.