An anomaly detection-based dynamic OD prediction framework for urban networks

The dynamic origin-destination (OD) information is crucial for traffic operations and control. This paper presents a dynamic traffic demand prediction framework based on an anomaly detection algorithm. The Principal Component Analysis (PCA) method is applied to extract main demand patterns which are used to detect the abnormal conditions. The proposed approach can select prediction methods (parametric or nonparametric) automatically based on the pattern detection results. Both simulation and field observed Automatic Number Plate Recognition (ANPR) data are used to verify the proposed approach where the Kalman filter model and the K-nearest neighbor model are chosen as the basic prediction methods. The results show that the prediction framework can effectively reduce the noise of a single prediction model particularly in the abnormal conditions and provide more accurate and reliable prediction results.