A Hybrid Markov–LogNormal Model for Traffic Scene Generation

The development and validation of advanced vehicle systems, such as energy management strategies for electrified vehicles, rely heavily on simulation-based testing. A key challenge lies in generating high-fidelity traffic scenes that accurately reflect real-world multi-modal conditions. Conventional data sources, such as standard driving cycles, lack the inherent stochasticity and multi-modal characteristics required for comprehensive system validation. To address this gap, this paper introduces a data-driven approach that integrates OpenStreetMap with real-time API data to construct a large-scale urban traffic dataset. Three typical driving conditions—"Urban Unimpeded,” “Urban Congested,” and “Highway/Mixed"—are identified using K-means clustering. A hybrid generation model is developed, combining a Markov Chain to simulate discrete “road type–congestion level” transitions and a bivariate log-normal joint distribution to model correlated continuous parameters, thereby preserving their intrinsic relationships. Experimental results demonstrate that the generated scenes exhibit both macroscopic statistical consistency and microscopic realism. These scenes can be directly imported into SUMO, providing valuable data support for vehicle system validation, including the development of energy management strategies for plug-in hybrid electric vehicles.