Complex electromechanical system abnormal state detection method based on multi-source data

Abstract

The invention discloses a complex electromechanical system anomaly detection method based on multi-source data, and the method comprises the steps: firstly carrying out the quantitative analysis of the correlation between multi-dimensional variables through employing detrended cross-correlation analysis (DCCA), and building a complex system multivariable coupling relation network taking the inter-variable coupling relation as an edge and a monitoring variable as a node; on the basis, establishing a variational diagram self-encoding model based on unsupervised learning; carrying out feature extraction on the multivariable coupling relation network of the system; training the model by using normal data, learning distribution of input data by using a graph convolution network as an encoder, performing sampling to obtain potential representation of the input data so as to realize reconstruction of a coupling network, obtaining a reconstruction probability threshold value by training a sample, and using a reconstruction probability as a system multi-dimensional polymorphic monitoring data anomaly detection evaluation index. According to the method, the coupling relation between the multi-source data is considered, the variational diagram auto-encoder model is introduced, the experience dependence is reduced, the problem of few abnormal samples is solved, and the accuracy and reliability of system anomaly detection are improved.

Description

technical field [0001] The invention relates to the technical field of abnormal detection of complex electromechanical systems, in particular to a method for detecting abnormalities of complex electromechanical systems based on multi-source monitoring data. Background technique [0002] The traditional anomaly detection of complex electromechanical systems mainly uses threshold alarms and human judgment, which is highly dependent on experience, low in accuracy, and high in misjudgment and missed detection rates. [0003] In recent years, with the increasing degree of automation and information integration of complex electromechanical equipment, the amount of data on the operating status of complex electromechanical systems has increased sharply. The data collected by representative monitoring systems such as DCS, TSI and SCADA equipped with system-level equipment have Features such as multiple sources, massive volume, abnormal shortage, and no label. Data-driven anomaly det...

AI Technical Summary

Problems solved by technology

[0004] At present, the anomaly detection methods based on deep learning do not consider the complex coupling relationship between the monitoring variables of the system, and are limited to the use of a single variable or multi-variable simple superposition for anomaly detection or fault identification. Reliability is difficult to meet the needs of industrial time series data anomaly detection

Images