Hierarchical time memory industrial anomaly detection method and device based on dimensionality reduction

Abstract

The invention proposes a method and device for industrial abnormality detection based on dimensionality reduction based on hierarchical time memory. The method includes: acquiring a data set corresponding to a multi-dimensional time series generated by an industrial sensor to be detected; denoising the original data; abstracting the vertices of dimension information into a corresponding number according to the number of dimensions in the data set; The correlation value is assigned to the corresponding edge in the graph as a weight value; the minimum spanning tree selection clustering is performed according to the correlation between dimensions; PCA is used to reduce the dimension of the formed block clusters; The layer time memory model judges abnormality. Through the method of the present invention to reduce the dimension through the selection of the correlation dimension, remove the redundant features, and reduce the amount of calculation, the time point anomaly in the time series data stream of the industrial sensor can be found in time, and the collection of normal industrial data sets for model analysis can be avoided. Cumbersome training.

Description

technical field [0001] The invention relates to the field of anomaly detection of industrial sensor multidimensional time data, in particular to a dimensionality reduction-based hierarchical time memory industrial anomaly detection method suitable for generating fast data streams in industry. Background technique [0002] In recent years, with the development of the Internet of Things and artificial intelligence, revolutionary changes have taken place in various fields. Industry 4.0 is one of them. Industry 4.0 is also known as industrial intelligent manufacturing. Designed for highly flexible automated production processes. Sensing devices on factory production lines are collecting more and more real data, creating a constant stream of data. Often such data streams can be represented by time series patterns, and most of them exist in the form of multidimensional time series. It is extremely important and meaningful to carry out data mining and effective data analysis on...

AI Technical Summary

Problems solved by technology

[0007] Purpose of the invention: Aiming at the characteristics of large data volume and strong dynamics of the data flow generated by industrial production sensors, there are problems such as high dimensionality and long model training time. The purpose of the invention is to provide a layered time memory industry based on dimensionality reduction An anomaly detection method and device do not require pre-training in advance, and can directly capture the context of the data for anomaly detection, thereby improving the speed and accuracy of anomaly detection

Images