Skew time series anomaly detection method based on cost-sensitive hybrid network

Abstract

The invention discloses a skew time series anomaly detection method based on a cost-sensitive hybrid network. The method comprises the following steps: firstly, establishing and training a cost-sensitive hybrid network model consisting of a deep convolutional neural network, a gated recursive network and a cost-sensitive loss function, wherein local features of a time sequence are learned througha deep convolutional neural network. Sequence characteristics of a time sequence are learned through a gated recursive network; then, the features are combined for classification; in the model training process, a cost sensitive loss function is used for measuring the similarity between an output result and a true value, then parameters of a network model are adjusted through a back propagation algorithm, and different penalty factors are used for punishing error detection of the network model for different numbers and types of samples. The method provided by the invention is simple, efficient,high in precision and relatively high in robustness, and has relatively high detection precision for both a skew time series data set and a non-skew time series data set.

Description

technical field [0001] The invention belongs to the technical field of time series data anomaly detection, and relates to a skewed time series anomaly detection method based on a cost-sensitive hybrid network. Background technique [0002] Skewed time series data refers to data sets in which the amount of sample data in different categories differs greatly. In practical applications, most of the time series data obtained by engineering measurement are within the normal range, with only a very small number of outliers, which is a typical skewed time series data set. In the binary classification problem, the result of the general classifier will be biased towards the normal class, and the false detection rate of the abnormal class is very high. But in practical applications, the minority class is often the focus of our attention, for example, in the fault detection of spacecraft, disease diagnosis in the medical field, and credit card fraud in the financial field. [0003] T...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a skewed time series anomaly detection method based on a cost-sensitive hybrid network, which solves the problem of low detection accuracy of minority samples in the skewed time series data set in the prior art, and because the existing The algorithm does not fully learn the features of minority samples, which leads to a serious decline in classification accuracy

Images