基于图像与电流特征的电熔镁炉欠烧工况半监督分类方法  

Semi-supervised Classification of Semi-molten Working Condition of Fused Magnesium Furnace Based on Image and Current Features

作　　者：卢绍文 温乙鑫 LU Shao-Wen;WEN Yi-Xin(State Key Laboratory of Synthetical Automation for Process Industries,Northeastern University,Shenyang 110819)

机构地区：[1]东北大学流程工业综合自动化国家重点实验室,沈阳110819

出　　处：《自动化学报》2021年第4期891-902,共12页Acta Automatica Sinica

基　　金：国家自然科学基金(61991404,61833004)资助。

摘　　要：针对电熔镁炉异常工况识别任务,在半监督学习框架下提出一种将电流与图像两类特征融合的解决方案.主要贡献为:使用多元图像分析(Multivariate image analysis,MIA)技术代替人眼,更为准确客观地对镁炉火焰进行特征提取;利用基于熵正则化(Entropy regularization,ER)的半监督学习框架,同时使用具有强互补性的生产图像与电流数据进行工况分类,从而弥补了基于单一特征分类的某些缺点;采用交叉熵方法(Cross-entropy method,CEM)优化分类器目标函数,较传统优化方法显著地提升了训练速度.通过仿真数据与公开数据集测试并讨论了本文算法的优势,并通过工业数据验证了所提方法的有效性、应用价值与良好的鲁棒性.Aiming at the task of identifying abnormal working conditions of fused magnesium furnace,this paper proposes a solution that combines the two types of features of current and image under the framework of semi-supervised learning.The main contributions of this paper are:Using multivariate image analysis(MIA)technology to replace the human eyes,and extracting features of magnesium furnace flames more accurately and objectively;using a semi-supervised learning framework based on entropy regularization(ER),and at the same time using strong complementary production images and current data to classify working conditions,thereby making up for some shortcomings in classification based on single feature;the cross-entropy method(CEM)is used to optimize the objective function of the classifier,which significantly improves the training speed compared with the traditional optimization method.The advantages of the algorithm in this paper are tested and discussed through simulation data and public data sets;and the effectiveness,application value and good robustness of the method proposed in this paper are verified through industrial data.

关 键 词：电熔镁炉 异常工况判别 多元图像分析 特征融合 基于熵的半监督学习 交叉熵方法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TF822[自动化与计算机技术—计算机科学与技术]