Researchers at Michigan Technological University (MTU) have developed a data-driven optimization modeling framework to reconstruct the location-duration-path choices for the missing observations from incomplete data from connected vehicles. A paper on the work is published in Transportation Research Part C: Emerging Technologies.

As the development of mobile sensing technologies, daily trajectory data have the potential to improve the observability of travel demand due to the details of location and timestamp in a transportation network. For example, connected vehicles provide high-frequency (e.g., 0.1 s) time and location information. This information contains individual’s dynamic choices of location-duration-path at a sub-second (i.e., 0.1 s) level over many days.

However, the incomplete raw trajectory data with missing information are difficult to be directly used for modeling travel demand dynamics. The objective of this paper is to develop data-driven optimization models to estimate the missing observations by exploiting information from the incomplete raw trajectory data over many days. The proposed models reconstruct the missing location-duration-path choices for individual connected vehicles. The results provide a supplement for developing and calibrating integrated travel demand and dynamic traffic assignment models.

—Zhao and Zhang (2019)

Kuilin Zhang, assistant professor of civil and environmental engineering and affiliated assistant professor of computer science at MTU, believes this approach will be a cost-effective way to allow transportation planners to do everything from make more effective traffic congestion mitigation strategies to know where to build wider or new roads.

In the future, we’re going to have more connected vehicles. If we fill in the missing parts of the data they’re providing, we can get complete activity and travel for individual drivers, and then this data can be used to know demand.

—Kuilin Zhang

Any vehicle with wireless access, such as cellular or Dedicated Short-Range Communication (DSRC) technologies, is considered connected. IHS Automotive expects that 152 million actively connected cars will be on roads around the world by 2025 and that the average car will produce up to 30 terabytes of data every day.

Connected vehicle trajectory data could be used to make travel predictions, but Zhang has found that there are enough gaps in the data that they can’t be used to make reliable predictions.

In this study, the researchers used two-months of connected vehicle data from 2,800 cars provided by the Safety Pilot Model Deployment Program in Ann Arbor, Michigan. From it, they created a data-driven optimization approach to reconstructing the missing location-duration-path choices those cars make.

They processed many-day raw trajectories, observed a set of historical choices of location-duration-path and identified missing observations in space and time dimensions. To improve computational efficiency, they applied data-driven network-time prisms that reduced the search space for the missing choices.

Then, they formulated Distributionally Robust Optimization (DRO) models with likelihood bounds—a special case of data-driven optimization models using phi-divergences (i.e., χ² distance), to reconstruct the missing choices. To solve the minimax programs of the DRO models while maintaining tractability, they reformulated and solve the equivalent dual problems of the DRO models based on the strong duality theory.

The reconstructed choices can be used to improve the validation and calibration of the models. The activity-based models of travel demand dynamics give more details to transportation planning organizations. Better estimation of travel demand, Zhang said, will also help reduce congestion, decrease emissions and save energy.

Zhang believes that the value of this activity-based model goes beyond just accuracy—it will save money. Local governments, which often buy information drawn from GPS on commercial vehicles from private companies, or rely on the National Household Travel Survey, which is expensive to conduct, and only presents information about a sliver of drivers instead of the whole, can use the models to know more and pay less to observe their municipality’s driving habits.

He also said that this kind of modeling will be especially important if more cities follow New York City’s lead and start implementing congestion pricing, which will use license plate readers to charge tolls to drivers during peak traffic times.

The next step in this research is for the model to be applied to existing connected vehicle testbeds in Florida, New York and Wyoming, and provide insights there on how to use connected vehicle data.

This is a big data era. In addition to the safety benefit of connected vehicle technology, high-frequency data generated from connected vehicles offer big data set for new mobility solutions.

—Kuilin Zhang

Resources

• Shuaidong Zhao, Kuilin Zhang (2019) “A distributionally robust optimization approach to reconstructing missing locations and paths using high-frequency trajectory data,” Transportation Research Part C: Emerging Technologies, Volume 102, Pages 316-335 doi: 10.1016/j.trc.2019.03.012