基于特征层密集连接与注意力机制的宽度学习系统及其在锌浮选过程的应用

Dense connection and attention-based broad learning system and its application to zinc flotation process

本文针对宽度学习存在计算消耗资源大、计算过程冗余数据较多等问题,提出了一种基于特征层密集连接与注意力机制的宽度学习系统(DCA-BLS),并利用其建立锌浮选过程快粗槽底流品位在线预测模型.首先将宽度学习系统的特征层不同窗口进行密集连接,引入弹性网络进行稀疏化处理,利用注意力机制处理特征节点,获得不同特征节点的权值,再将加权后的特征节点与输入的数据相结合,共同作为增强层节点的输入,使模型更为紧凑.在公共数据集和锌泡沫浮选数据上将DCA-BLS与其他预测算法进行了对比实验,实验结果表明,本文提出的方法训练时间短,且相较于其他所比较方法具有更高的准确率.

In this paper,a dense connection and attention-based broad learning system(DCA-BLS)is proposed to address the issues of large computing resource consumption and redundant data in the training process of broad learning system.It is applied to predict the zinc ore froth flotation process fast coarse trough bottom flow grade.Firstly,the different windows in feature layer of the broad learning system are densely connected.Then elastic network is employed to sparse the model,and attention mechanism is proposed to deal with nodes in feature layer,to obtain the weights of different characteristics of the nodes.The weighted nodes in feature layer of BLS are combined with the input data.The combination is used as the input of enhance nodes in BLS,which can make learning model more compacted.DCA-BLS is compared with other prediction algorithms on public data sets and zinc froth flotation data.Experimental results show that the proposed method has shorter training time and higher accuracy than other methods.