基于特征融合宽度学习系统的锌浮选过程工况识别

Working Condition Recognition in a Zinc Flotation Process Based on Feature Fusion Width Learning System

浮选工况是浮选操作的重要判断依据,如何准确地识别浮选工况对浮选性能的提升有重要意义。基于机器视觉方法是浮选工况识别的主流方法,通常采用大数据技术在浮选工况数据集上建立浮选表层泡沫特征与浮选工况之间的关系模型,工况识别效果与工况数据集密切相关。一旦出现数据集中未包含的新工况,难以获得满意的识别效果。为此,针对当前大部分工况识别方法自适应性不足的问题,以锌精选为例,提出一种基于多特征宽度学习的锌浮选工况识别方法,以增量学习方式自适应新出现的工况。首先,根据多特征的不同特性,构建基于多特征宽度学习的锌精选工况识别模型;然后,在浮选状态变化和精选槽故障导致模型识别准确率降低时,通过拓宽特征层、增强层以及输出层的方式调整网络结构以进行增量学习。试验结果表明,基于多特征宽度学习系统的锌浮选工况识别方法具有良好的工况自适应性能,应用价值良好。

Flotation condition is an important basis for judging flotation operation,and how to accurately identify the flotation conditions is of great significance to the improvement of flotation performance.Machine vision-based method is the mainstream method for working condition recognition in froth flotation,and it usually construct the relationship model between the visual features of froth appearance and working conditions by big data technology on the flotation working condition dataset.The performance of working condition recognition is closely related to the working condition dataset.When the dataset doesn't include the new working condition,it is difficult to obtain satisfactory results.Therefore,a working condition recognition method based on feature fusion width learning system is proposed in a zinc flotation process.The proposed method can adapt to new working condition by incremental learning.Firstly,a working condition model based on the feature fusion width learning system is introduced according to different characteristics of multiple visual features.Then,when the accuracy of recognition model is reduced due to the change of working condition or device fault,the network structure is adjusted by widening the feature layer,enhancement layer and output layer to implement incremental learning.The experimental results show that the proposed method based on feature fusion width learning system has nice robustness and self-adaptive performance in the real application.