Spatiotemporal Modeling and Real-Time Prediction of Origin-Destination Traffic Demand

Traffic demand prediction has been a crucial problem for the planning, scheduling, and optimization in transportation management. The prediction of traffic demand counts for origin-destination (OD) pairs has been considered challenging due to the high variability and complicated spatiotemporal correlations in the data. Though several articles have considered estimating traffic flows from counts observed at specific locations, existing traffic prediction models seldom dealt with spatiotemporal demand count data of certain OD pairs, or they failed to effectively consider domain knowledge of the traffic network to enhance the prediction accuracy of traffic demand. To tackle the aforementioned challenges, we formulate and propose a multivariate Poisson log-normal model with specific parameterization tailored to the traffic demand problem, which captures the spatiotemporal correlations of the traffic demand across different routes and epochs, and automatically clusters the routes based on the demand correlations. The model is further estimated using an expectation-maximization algorithm and applied for predicting future demand counts at the subsequent epochs. The estimation and prediction procedures incorporate Markov chain Monte Carlo sampling to overcome the computational challenges. Simulations as well as a real application on a New York yellow taxi data are performed to demonstrate the applicability and effectiveness of the proposed method. Supplementary materials for this article are available online.