Transformer multivariate time series anomaly detection based on GAT
A multi variable time series anomaly detection method based on GAT Transformer is proposed to address the complex dependency relationships between multiple times and variables in multidimensional temporal data,which cannot accurately identify a small number of outliers.Firstly,convert the features into embedded vector representations;Then,the graph attention mechanism is introduced to adaptively learn the complex dependency relationships between different times and variables;Finally,concatenate the original data with the output of the GAT layer,input it into a Transformer encoder with positional encoding,and determine the anomaly situation by calculating the anomaly score and setting a threshold.The results indicate that the proposed model can effectively detect anomalies in temporal data.
anomaly detectiongraph attention networkTransformermultivariate time series
NETL
NSTL
万方数据
