基于半监督聚类方法的管道运行状态识别研究  

Research on pipeline operation status identification based on semi-supervised clustering method

作　　者：方明月 冯早 朱雪峰 Fang Mingyue;eng Zao;Zhu Xuefeng(School of Information Engineering&.Automation,Kunming University of Science and Technology,Kunming,650500,China;Mineral Pipeline Engineering Technology Research Center of Yunnan,Kunming,650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,昆明650500 [2]云南省矿物管道输送工程技术研究中心,昆明650500

出　　处：《南京大学学报（自然科学版）》2023年第3期435-445,共11页Journal of Nanjing University（Natural Science）

基　　金：国家自然科学基金(61563024)。

摘　　要：目前工业现场采集的大部分数据缺失标记信息,若仅使用其中的标记数据和未标记数据的局部信息会造成学习器的性能下降,而传统半监督学习对标记数据的利用具有随机性且没有考虑监督信息数量的变化对性能的影响.针对上述问题,提出一种主动学习策略下基于最小生成树的变分贝叶斯推理半监督高斯混合模型,利用主动学习策略在标记信息中重新筛选构建高质量成对约束信息,利用最小生成树聚类来初始化模型参数,提升模型在标记样本数量受扰动时的鲁棒性.选用实验室采集的管道状态检测数据集对提出的模型进行验证,实验结果表明,当标记样本占总体样本的比例从50%下降到10%时,提出的组合模型的预测准确率依旧保持在72.4%以上,而且,当某一类别的样本完全失去监督信息时,该组合模型的聚类效果仍然可以分析判断数据类别的所属类型.Most of the data in current industrial field data collection are unlabeled,and the performance of the learner is degraded if only the labeled data in the dataset and the local information of the unlabeled data are used.Traditional semisupervised learning uses the labeled data randomly and does not consider the impact on the performance when the number of supervised information changes.To address these problems,this paper proposes a semi-supervised Gaussian mixture model based on minimum spanning tree with an active learning strategy for variational Bayesian inference.In the model,the active learning strategy is used to re-screen and construct high-quality pairwise constraint information among the labeled information,and the minimum spanning tree clustering is used to initialize the parameters of the model and improve the robustness of the model under the perturbation of the number of labeled samples.Experimental results show that the prediction accuracy of the proposed combined model remains above 72.4%when the proportion of labeled samples decreases from 50%to 10%of the total samples.The clustering effect of the combined model is stll able to analyze the data category when a category completely loses the supervisory information.

关 键 词：半监督聚类 主动学习策略 高斯混合模型 成对约束 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]