Lightweight unsupervised anomaly detection method based on multivariate time series data analysis

Abstract

The invention discloses a lightweight unsupervised anomaly detection method based on multivariate time series data analysis. The method comprises two models: a detection model and an inference model; the detection model firstly extracts time dependence characteristics of the captured multivariate time series data through a random convolutional neural network, and then encodes and decodes the multivariate time series data after the characteristics are extracted by using a deep Bayesian network, and the detection model can determine a detection precision range; the inference model is composed of a score attention unit, a threshold value automatic selection unit and a point adjustment unit, the score attention unit adopts an attention mechanism to expand the feature difference between abnormal data and normal data and provides a theoretical basis for abnormal interpretation, and the threshold value automatic selection unit can automatically calculate a threshold value, the point adjustment unit can simulate the generation process of real anomalies, and the inference model can improve the accuracy, stability and interpretability of anomaly detection. According to the invention, the method can cope with rapidly increased data scale and complex and changeable exception types.

Description

technical field [0001] The invention belongs to the field of machine learning anomaly detection, in particular to a lightweight unsupervised anomaly detection method based on multivariate time series data analysis. Background technique [0002] Anomalies are observation individuals or subsets that differ significantly from the ontology of observations, and these individuals or subsets deviate from the original generation pattern. Anomaly detection is the process of finding these individuals or subsets. Multivariate time-series data is a collection of numerical sequences with a sequential relationship. In particular, each element in the numerical sequence is a multidimensional vector; multivariate time-series data can describe the state of the observed value ontology and implicitly indicate the change law of the phenomenon, so the analysis Multivariate time series data is very important in the field of anomaly detection. [0003] Traditional supervised learning methods requ...

AI Technical Summary

Problems solved by technology

Compared with OmniAnomaly, this method has a great improvement in training time, but its inherent problems such as mode collapse, difficulty in convergence, and difficulty in training make its practicality and stability greatly reduced, so it is not easy to use in practice

Images