Traffic Anomaly Prediction based on Spatio-Temporal Uncertainty

Traffic Anomaly Prediction (TAP) has emerged as a critical concern in the field of Intelligent Transportation Systems (ITS), attracting attention from numerous scholars. Accurately predicting traffic anomalies is challenging due to the inherent uncertainty in their occurrence. Unfortunately, most existing methods overlook this issue. They rely solely on deterministic modeling approaches. Moreover, many TAP methods are only applicable to simple transportation scenarios, disregarding the complex spatio-temporal dependencies of traffic anomalies. To address these challenges, we propose a novel TAP method that integrates deterministic and uncertainty modeling approaches. Firstly, we design an attention-based temporal convolutional module to capture temporal dependence from anomaly data. Secondly, we introduce a diffusion graph convolution module to capture spatial dependence across different regions. These two modules are then combined to extract spatio-temporal features of the anomaly pattern. Subsequently, we employ a probabilistic module to address potential uncertainty. Finally, we develop a model for TAP based on spatio-temporal uncertainty. We collected and processed traffic anomaly data from 24 regions in Beijing. Experimental results demonstrate that our proposed method outperforms existing models.