Transportation Domain Knowledge-based AI Applications
AI and data acquisition technologies have been broadly applied for planning and managing transportation networks in recent decades. However, we still face the challenge of identifying appropriate application scenarios of AI in transportation and developing effective solutions. Seizing this opportunity, the AI+CASE Lab works on integrating AI and transportation domain knowledge to explore latent features of the transportation system and develop physics-informed, interpretable, and robust solutions to address traffic problems (Figure 5). Supported by NSF and DOT UTC projects, the aggregated effort of our lab made the following significant contributions:
- Developed transportation domain knowledge-based data-driven/AI approaches (e.g., deep learning, reinforcement learning) for traffic speed, shockwave, traffic events, and traffic network performance detection and prediction. The proposed approaches leverage traffic flow theory and AI technologies to capture traffic dynamics in real-time.
- Integrated the domain knowledge and ML/AI approaches (Gaussian Process Regression, community learning, clustering, regression) into large-scale optimization modeling and solution approaches to address connected and autonomous vehicle control and operation problems. The proposed approaches can efficiently solve the optimization model online by reducing the solution space, which is achieved by capturing the traffic features through AI technologies.
Figure 5. The framework of transportation domain knowledge-based AI.
Funding source: STRIDE UTC F2 project; NSF award CMMI 2213459; NSF career award; NSF SCC-PG.