Fusion of Simulation and AI Methods for Understanding HOV/HOT Lane Operational Flow Dynamics

This study investigated the impact of converting High Occupancy Vehicle (HOV) lanes to High Occupancy Toll (HOT) lanes on fundamental traffic flow characteristics, focusing on speed, density, and flow relationships. A 25-mile HOV corridor along I-24 Westbound in Nashville, Tennessee was evaluated using both microscopic simulation via VISSIM and data-driven machine learning through a Multi-Layer Perceptron (MLP) neural network. Four operational scenarios were assessed: (1) HOV lanes without enforcement, (2) HOV lanes with effective occupancy enforcement, (3) HOT lanes with limited access points, and (4) HOT lanes with intermediate access points. Flow-density and speed-flow relationships were modeled using Greenshields theory to extract key traffic performance thresholds including free-flow speed, jam density, and maximum flow. Results indicate that while free-flow speeds were generally consistent across scenarios (ranging from 71 to 80 mph), HOV and HOT lanes exhibited higher values compared to general-purpose lanes. Capacity increases were observed following HOV-to-HOT conversions, especially when intermediate access points were introduced. The MLP neural network successfully replicated nonlinear flow relationships and predicted maximum flow near 2000 vph with a jam density of approximately 215 vpmpl—values that closely matched simulation outputs. Both the VISSIM and MLP-derived diagrams demonstrated curve shapes and capacity thresholds that were highly consistent with Highway Capacity Manual (HCM) standards for freeway segments. However, slightly higher thresholds were observed for HOV/HOT lanes, suggesting their potential for improved operational performance under managed conditions. The integration of simulation and machine learning offers a robust framework for evaluating managed lane conversions and informing data-driven policy. Beyond the scenario-specific findings, the study demonstrates an innovative hybrid methodology that links detailed microsimulation with an explainable neural network model, providing a concise and scalable approach for analyzing managed-lane operations. This combined framework highlights the contribution of integrating simulation and AI to enhance the analytical depth and practical relevance of traffic flow studies.