Research On Anomaly Detection Of Industrial Times Series Data Based On Unsupervised And Semi-supervised Learning

The rise of Big Data has brought various challenges and revolutions to many fields.Even though its development in many industries has gradually become perfect or even mature,its application and development in complex industrial scenarios is still in its infancy.In recent years,the incompatibility between the incompleteness of industrial big data anomaly detection research and the increasing demand for intelligent manufacturing has become a serious bottleneck problem.Meanwhile,many industrial manufacturing practitioners have a blindly optimistic understanding of abnormal pattern detection technology,which leads to abnormal operation of many equipment.if couldn't be handled in time,it will affect production safety and even cause serious losses in financial,manpower and material resources.Based on the above background,this paper will make an in-depth research on the anomaly detection problem of industrial time series data.Our research has the following two aspects:(1)Research on single-dimensional time series point anomaly detection based on Unsupervised Learning: Unlike periodic time series,aperiodic or weakly periodic time series in industrial scenarios are more common.Considering the need for online real-time monitoring,we need to solve the problem of point anomaly detection of oil chromatographic characteristic gases.Thus,we propose a sliding window-based method for unsupervised single-dimensional time series point anomaly detection problem which called Confidence Interval Radius Slope Method(CIRS).CIRS is a fusion of knowledge-driven and data-driven to realize online real-time monitoring of possible data quality problems.(2)Research on multi-dimensional time series pattern anomaly detection based on Semi-supervised Learning: Considering the difficulty of manual labeling data in complex industrial scenes,we need to solve the problem of transformer fault pattern anomaly detection.Thus,we propose a semi-supervised multi-dimensional time series vector classification algorithm,which called Self-training Repeat-marker Method(SRM).SRM improves the utilization of unlabeled data,enhances the classification ability of the base classifier and realizes the judgment of the specific fault type in transformer.From the experimental results,CIRS has obtained higher PR values than other unsupervised methods by the subject data.And meanwhile,SRM obtained a higher Kappa coefficient than the traditional self-training method.