Efficient Data Management for Intelligent Urban Mobility Systems

Transit agencies increasingly need to leverage large-scale multimodal sensor data to optimize operations and support decision-making, but lack standardized frameworks for collecting, storing, and processing such data efficiently. Traditional approaches often fail to handle the high-volume, heterogeneous nature of real-world transit data streams. This work is innovative because it provides a practical, cloud-based architecture that enables transit agencies to build machine learning applications using real-world data while maintaining the flexibility to adapt systems as new requirements emerge.

This paper develops efficient data management and processing frameworks for intelligent urban mobility systems using real-world transit data from Chattanooga. The work presents an integrated data architecture combining real-time vehicle telemetry, weather data, traffic information, and elevation maps to support machine learning applications. The framework addresses challenges in storing high-velocity data streams and enabling both offline model training and real-time inference for transit applications including energy prediction and ridership forecasting.

The framework successfully demonstrates integration of multiple data sources into a unified system supporting both offline training and real-time inference. Results show that the architecture enables effective machine learning-based energy prediction for electric vehicles and transit optimization applications. The system has been deployed with the Chattanooga Area Regional Transportation Authority and provides the foundation for multiple AI applications including occupancy prediction and route-level energy consumption estimation.