Improving Metropolitan-Scale Transportation Systems
with Data-Driven Cyber-Control

NSF CNS-1446640
                    

                                                Real-world Impact

New York City Transit System

We have initiated a collaboration with Transit App, which works with 112 metropolitan areas and improves almost 2 billion trips with an integrated transit service, e.g., bus arrival prediction and trip planning including Uber, car2go, and bikeshare. We aim to use taxi, bus and subway network data along with trip plan data to model human mobility for New York City from a multi-modality perspective. Different from existing human mobility modeling based on single modalities, this multi-modality approach has a potential to address the overfitting challenge, and provides more accurate analyses. Based on this multi-modality human mobility model, several services, such as real-time ridesharing, can be designed to improve the efficiency of urban transportation in New York City.



Washington D.C. Bikeshare Network

We have initiated a collaboration with Department of Transportation in Washington D.C. to improve the efficiency of Capital Bikeshare Network. In particular, we aim to design an efficient bike rebalancing algorithm for rebalancing trucks. In this DC station map, the bigger the dots, the higher the demand dynamics. In a time slot, some stations do not have any bikes, while some stations do not have any docks. This rebalancing algorithm is challenging because passenger demand is dynamic and data we have are incomplete. Thus, we aim to propose uncertain demand and supply models for D.C. Bikeshare Network, and then apply a model predictive control to optimize the rebalancing process by minimizing rebalancing costs and supply/demand mismatch.



Shenzhen Bus Network

We have been working with SIAT to design and implement a bus arrival prediction app for Shenzhen. The preliminary version of app has been used by more than 100,000 users and their request counts as given in the figure. But the key challenge we have for this service is that Bus GPS data are uncertain, e.g., they are incomplete, untimely, and inconsistent with metadata. Thus, we propose a solution to integrate multi-source data from nearby taxis and smartcards to increase the accuracy of bus arrival prediciton. In particular, we aim to design a data fusion technique to recover uncertain GPS data based on context-aware tensor decomposition to minimize recovering errors.