Prediction of Next-Time Traffic Congestion With Consideration of Congestion Propagation Patterns and Co-Occurrence

Traffic congestion on urban roads has become one of the most serious concerns in cities as it results in considerable economic loss or even accidents. A proper determination and prediction of traffic congestion in urban transport networks allows timely implementation of interventions whilst allowing the understanding on how traffic congestion occurs and evolves. This study investigates the propagation patterns of traffic congestion and proposes a probability model to predict the next-time congestion within an urban road network. The study presents the means to identify traffic congestion, followed by a traffic congestion prediction model with consideration of temporal and propagation patterns. Markov Random Field (MRF) is adopted to construct congestion areas – subset of congested spatial grids to ensure spatial correlations. Grid vectors are generated with word embedding methods to consider the co-occurrence of congested grids and thereafter to present propagation patterns and temporal characteristics with congestion context vectors and temporal congestion vectors, respectively. A probability-based traffic congestion prediction model is formulated with a collaborative filtering approach to represent the congestion probability of each spatial grid with the grid vectors. A case study of Singapore with a trajectory dataset over a month demonstrates that the developed model provides good performance in traffic congestion propagation patterns and congestion prediction. The proposed methods provide an accessible mean to model congestion propagation patterns and predict congestion, which could be a generalized approach for urban areas.