基于多元时间序列特征解释CNN的阳极效应识别方法  

Anode effect recognition method based on feature explanation CNN of multivariate time series

作　　者：常玉杰 邓紫晴 陈晓方[2] 邹忠 Chang Yujie;Deng Ziqing;Chen Xiaofang;Zou Zhong(Qingtongxia Branch of Qingtongxia Aluminum Co.,Ltd.,Wuzhong 751603,China;School of Automation,Central South University,Changsha 410083,China;School of Metallurgy and Environment,Central South University,Changsha 410083,China)

机构地区：[1]青铜峡铝业股份有限公司青铜峡分公司,宁夏吴忠751603 [2]中南大学自动化学院,湖南长沙410083 [3]中南大学冶金与环境学院,湖南长沙410083

出　　处：《轻金属》2023年第12期18-24,共7页Light Metals

摘　　要：阳极效应的识别对于保证工业铝电解槽稳定运行和精细化控制具有重要意义。针对阳极分布电流数据高维、变量之间相互影响的特点,以及提高卷积神经网络(CNN)可解释性会影响模型性能的问题,提出一种用于阳极效应识别的多元时间序列特征解释CNN模型。与先进的多元时间序列分类器对比的实验结果表明,该模型能够达到较高的阳极效应识别准确率。The anode effect recognition is of great significance to maintain the stable operation and fine control of the aluminum electrolytic cell.To address the problem of convolutional neural network(CNN)with improved explanation affecting model performance,a feature explanation CNN model of multivariate time series is proposed for anode effect recognition,which aims at the data characteristics of the high dimensionality and the mutual influence among variables of the anode distribution current.Experimental results show that the proposed model can achieve higher accuracy in anode effect recognition,compared with advanced multivariate time series classifiers.

关 键 词：工业铝电解 多元时间序列 特征解释 卷积神经网络 阳极效应识别 

分 类 号：TF821[冶金工程—有色金属冶金]