Deep Learning Hybrid Model for Equitable Data Representation: Physics Model Incorporating Ramp Traffic Dynamics & Data-Driven Enhancements

Sparse sensor deployment and sensor failures are persistent traffic monitoring and management challenges, leading to critical network dynamics being neglected or overlooked. This study introduces the Fundamentally-Informed Ramp Adjustment Model (F-RAM), a novel hybrid framework for generating robust and reliable traffic flow data in underrepresented regions to address these challenges and support equitable infrastructure planning. F-RAM combines the Cell Transmission Model (CTM), an established physics-based traffic model, with data-driven refinements to capture localized dynamics and ramp-specific flow adjustments typically uncaptured by traditional physics methods. By incorporating ramp contextual features and intermediate sensor data, F-RAM adapts the generalized fundamental traffic relationship within the CTM to account for the unique characteristics of sparsely sensed environments, such as unobserved ramps. Validated on real-world highway segments with transitional point-based sensor data and ramps, F-RAM demonstrates its ability to refine sparse data representations by improving the standalone CTM's traffic state estimation (TSE) accuracy by an average of 44 % in our experimental simulations. This hybrid framework bridges fundamental traffic flow principles with modern computational tools, offering actionable insights into localized traffic dynamics and supporting equitable, data-driven infrastructure planning. Additionally, the methodology lays a foundation for future integration with emerging connected and autonomous vehicle (CAV) technologies, highlighting its potential adaptability to dynamic, data-rich environments.