Analytics on Bigger Data in Traffic management: Spatiotemporal Origin-Destination Prediction

Significance 

Traffic has remained a major global problem in contemporary society. With technological advances and the development of intelligent transport systems (ITPs), traffic data collection is made possible through auxiliary systems such as cameras and a global positioning system. The availability of traffic data provides more opportunities for traffic prediction, which plays a significant role in transportation management. In particular, the online and automatic prediction of origin-destination demand in traffic systems has recently attracted considerable research attention owing to its growing impact on traffic management policies. The origin-demand count represents the number of trips made between the origin and a destination of a particular transport network. It has proved useful in planning charging services of electric vehicles and meeting travel demands by removing empty vehicles on the road.

Traffic prediction has been extensively studied. Unfortunately, these studies fail to consider the OD demand count data and the stochastic nature of the traffic demands. This presents a research gap that needs to be addressed urgently. “The extension from location-based traffic prediction to OD prediction is not trivial due to dimension increase and the much more complex dependence structure, said Dr. Xiaochen Xian, assistant professor at the University of Florida. The difficulty in modeling and predicting OD traffic demand counts can be attributed to several challenges. They include the computational issues due to the complicated spatiotemporal correlations and high variability in the collected data. Moreover, most traffic prediction models do not consider the spatiotemporal demand count of OD pairs which may lead to inaccurate predictions.

From the challenges highlighted, it is evident that incorporating the physical domain knowledge of the traffic networks and the sparsity of the structural correlation is promising for improving the traffic demand prediction accuracy. Equipped with this knowledge, Dr. Xiaochen Xian from the University of Florida together with Mr. Honghan Ye, Dr. Xin Wang and Dr. Kaibo Liu from the University of Wisconsin-Madison developed a spatiotemporal and real-time model for predicting the origin-destination traffic demand. The aim was to overcome the problems mentioned above to improve the robustness and accuracy of the online traffic demand prediction for efficient transportation management. Their work is published in the journal, Technometrics.

In their approach, a multivariable Poisson log-normal model was formulated and investigated. The model comprised specific parametrization tailored to account for the spatiotemporal correlations for different routes and epochs within the traffic network. Furthermore, the authors used an expectation-maximization algorithm to estimate the model and predict the demand counts for additional epochs. Additionally, the Markov chain Monte Carlo sampling was included to alleviate the computational issues. Finally, the applicability of this method was verified through simulations and real-time applications on New York taxi data.

Results showed that the proposed model allowed for full exploration of the complex spatiotemporal structural correlation of the traffic network demand allowing for the automatic clustering of the routes with high correlation. Furthermore, the model parameters, estimated with high accuracy, were successful for online traffic prediction. The improved performance could be attributed to the domain knowledge integration and reduced computational burden. Furthermore, the feasibility of the proposed model was validated both numerically and through the real application of the New York traffic data.

In summary, a multivariable Poison-log normal model was reported to solve the inherent challenges in modelling and prediction of traffic demand counts. Due to the specialized parametrizing, the model exhibited improved estimation and superior estimation performance than the existing models. Based on the results, scalability was necessary to enhance the method’s applicability for large and complex traffic networks. Interestingly, the proposed method is versatile and can be applied in other networks with count data with minimal modifications. In a statement to Advances in Engineering, Dr. Xiaochen Xian, first author explained that it would improve modeling and prediction of different networks with count data such as transportation systems.

Analytics on Bigger Data in Traffic management: Spatiotemporal Origin-Destination Prediction - Advances in Engineering

About the author

Dr. Xiaochen Xian is an assistant professor from the Department of Industrial and Systems Engineering at the University of Florida.

.

.

About the author

Honghan Ye is a Ph.D. candidate from the Department of Industrial and Systems Engineering at UW-Madison.

.

.

About the author

Dr. Xin Wang is an assistant professor from the Department of Industrial and Systems Engineering at UW-Madison.

.

.

About the author

Dr. Kaibo Liu is an associate professor from the Department of Industrial and Systems Engineering at UW-Madison.

.

.

Reference

Xian, X., Ye, H., Wang, X., & Liu, K. (2020). Spatiotemporal Modeling and Real-Time Prediction of Origin-Destination Traffic DemandTechnometrics, 63(1), 77-89.

Go To Technometrics

Check Also

How can we estimate geo-properties in deep zones on a regional scale? - Advances in Engineering

How can we estimate geo-properties in deep zones on a regional scale?