By 2030, nearly 146 million connected vehicles will be in operation in the U.S. Each vehicle will generate 25 gigabytes of data per hour; however, traditional data analytics systems are not capable of handling the high amount of data. This will require the development of novel computational and scientific tools to shape our future connected vehicle cyberinfrastructure. The project aims to create a new data mining engine embedded at the vehicle level to collect and process the streams of data in a highly efficient and scalable way. Harnessing novel data science tools, this project will address important fundamental questions of how to efficiently decide what information to collect, what to filter, and what to process within a vehicle. In turn, having an efficient in-vehicle data collection engine will improve our capabilities to build a high-impact cyberinfrastructure for connected vehicles. This cyberinfrastructure will open up new research capabilities and provide social benefits in the fields of intelligent transportation and smart cities. The data will tell us how to improve traffic congestion, how to create safer streets, and how to better design our cities, resulting in improved mobility, economic improvements, and lives being saved. This project embraces education and workforce development to educate our future data scientists and engineers through the creation of interdisciplinary learning environments, courses, and research-intensive programs. This project will also broadly promote undergraduate research, K-12 educational outreach activities, and STEM education in underrepresented groups.
This project will investigate novel approaches to understand connected vehicle data, comprised of spatio-temporal trajectories and non-spatial sensor data, and develop scientific tools that can compress, impute, partition, and summarize connected vehicle data, while addressing important data challenges and scalability issues associated with the large scale of the database. First, the data collection platform will minimize data redundancy at a vehicle level through a proximity-based data sampling mechanism. Second, the platform will improve the integrability of connected vehicle data by capturing map-consistent trajectory points. Third, it will link dispersed connected vehicle data with map data by harnessing the special characteristics of new trajectory data formats. This scientific investigation centers on matrix decomposition, optimization, and spectral clustering techniques and includes city-wide vehicle sensor deployment to validate the proposed effort based on real-world and large-scale connected vehicle data. Results of the project will challenge the scientific gap between trajectory data analytics and the matrix decomposition techniques. Finally, this project will advance scientific knowledge in 1) trajectory data compression based on non-spatial sensor data, 2) the matrix decomposition techniques for the location inference of connected vehicle data, 3) the spectral graph partitioning methods for trajectories and non-spatial sensor data, and 4) large-scale trajectory data mining.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
