MTGAE: Graph Autoencoder with Mirror TCN for Traffic Anomaly Detection

Abstract

Traffic time series anomaly detection has been intensively studied for years because of its potential applications in intelligent transportation. Classical traffic anomaly detection methods ignore the hidden dynamic associations between road network nodes as it evolves, resulting in the inability to capture the long-term temporal correlation, spatial and temporal characteristics of traffic data, and the abnormal nodes from it in a dataset with high periodicity and trends (e.g., morning peak travel). In this paper, we proposed a Mirror Temporal Graph Autoencoder (MTGAE) framework to explore anomalies and capture unseen nodes and the spatiotemporal correlation between nodes in the traffic network. In the method, we propose Mirror Temporal Convolutional Module (MTCM) module, to our best knowledge, which possesses better feature extraction and captures hidden node-to-node features in the traffic network and will be passed as hidden to the next module. Our proposed Graph Convolutional Gate Recurrent Unit Cell (GCGRU CELL) module, mapped to a high-dimensional space by Gaussian kernel functions, which discerns anomalous information in the traffic network and possible anomalies in the complex interdependencies between nodes and nodes based on prior knowledge and input data. We compared our work with several other advanced deep-learning anomaly detection models. The experimental results on the NYC dataset prove that our model works best compared to other models for traffic anomaly detection.