Unveiling Multistability in Urban Traffic Through Percolation Theory and Network Analysis

Traffic congestion poses a persistent challenge for modern cities, yet the complex behavior of urban road networks—particularly multistability in traffic flow—remains poorly understood. To address this gap, we analyzed a high-resolution traffic dataset from four Chinese cities over 20 working days (5-min intervals), applying percolation theory to characterize system performance via congestion rate (f) and the size of the largest functional cluster (G). Our analysis revealed clear bimodal and multimodal distributions of G versus f across different periods, ruling out random failure models and confirming the presence of multistability. Leveraging data-driven clustering and classification techniques, we demonstrated that road segments with high betweenness centrality are disproportionately likely to become congested, and that the top 1% most topologically important roads accurately predict both stable state types and the joint behavior of G and f. These findings offer the first large-scale empirical evidence of multistability in urban traffic, laying a quantitative foundation for forecasting phase transitions in congestion and informing more effective traffic management strategies.