Congestion Pattern Prediction for a Busy Traffic Zone Based on the Hidden Markov Model

Abstract

Congestion pattern in a busy traffic zone would not always be a single state. The scenario of congestion occurring in multiple connected road links may be shifted to the scenario of whole zone network paralyzation immediately. It is necessary to dynamically predict the congestion pattern for a busy traffic zone rather than a road link, which could provide information for network system decision making. Considering the close connection between the upstream and downstream traffic flow, this study proposes a congestion pattern prediction model for a busy traffic zone based on the hidden Markov model (HMM). The model establishes a correlation between the external road traffic state (observation state) and internal road traffic state (hidden state) of a busy traffic zone. We acquire these traffic states by cleaning and mining floating vehicle trajectory data. With these data, we calibrate the HMM and predict the zone congestion pattern. This article demonstrates the validity and rationality of the model by taking a hospital area in Ningbo City as an example. The prediction accuracy can reach 83.4%, which is 5.8% higher than that of the autoregressive moving average model. This case result illustrates the feasibility and effectiveness of our approach in the field of congestion pattern prediction for busy traffic zone.

Keywords

• Regional congestion
• hidden Markov model
• congestion pattern