Research on network-level traffic pattern recognition

Abstract

Real-time network-level signal control, traffic assignment and route guidance are promising approaches for alleviating congestion. Different optimal sets of control parameters and strategies for area-wide signal control, traffic assignment and route guidance can be determined according to different traffic patterns using many methods. Because of the importance of pattern recognition of network-level traffic patterns in traffic control and other applications, we present some elementary research on the topic based on the theories and methods of pattern recognition. First, we formulate the general process of network-level traffic pattern recognition, then some useful methods, such as PCA and SVM, are used for feature extraction, training and classifying of network-level traffic patterns. The experimental results show that the effectiveness of the proposed methods.