Robustness anomaly detection method based on local and global statistical analysis

Abstract

The invention discloses a robustness anomaly detection method based on local and global statistical analysis. According to the method, modeling is carried out on the basis of local information and global information, then an comprehensive index is judged, and if the index exceeds a preset judgment condition, it is determined that an anomaly happens once. According to the method, an anomaly detection model in current industrial application is improved, a local model and a global model are combined, and the comprehensive index is comprehensively considered for early warning so that possible system faults can be avoided, and the effect of predictive maintenance is achieved.

Description

technical field [0001] The invention relates to the technical field of industrial equipment detection, in particular to a robust anomaly detection method based on local and global statistical analysis. Background technique [0002] Anomaly detection belongs to a large class of algorithms in unsupervised learning. Unsupervised learning is a learning method in which samples are not labeled. The purpose of anomaly detection is to detect data that deviates from the main data group in the data. These deviated data may indicate that there is an abnormal state of the equipment, such as abnormal operation or damage, etc., and the system can then give an early warning to avoid possible system failures and play a role in predictive maintenance. [0003] Commonly used anomaly detection algorithms include One-Class SVM, LOF (Local Outlier Factor), Isolation Forest, DBSCAN, etc. These algorithms are basically based on assumptions of statistical distributions or estimates of densities....

AI Technical Summary

Problems solved by technology

The data collected at the beginning has no abnormal labels, so this is an unsupervised learning problem

Images