Modeling lane-specific breakdown probabilities at freeway diverge sections

Highway capacity has a stochastic nature. Most of the previous studies estimated cross-section-based capacity distributions, which are not able to assess the probability of semi-congested states, where traffic breaks down on certain lanes while flows uninterruptedly on others. This study aims to explore the heterogeneity of lane-specific breakdown probabilities using statistical models. We selected six diverge sections of freeways in California for this study. Semi-congestion is a common phenomenon observed in diverge sections and thus it is important to estimate the lane-specific breakdown probabilities for more effective traffic management. A new method was proposed to identify optimal threshold speed for each lane by maximizing the reduction of traffic efficiency. A total of 4,463 lane-based breakdowns were identified at the selected sections based on the optimal threshold speeds. Log-rank and Wilcoxon tests provided strong evidence for the heterogeneity of capacity distributions among individual lanes of the same section and showed the necessity of modeling capacity distributions separately at the lane level. A Bayesian hierarchical Weibull model was developed to estimate lane-specific capacity distributions, which allowed model parameters to vary across freeways to account for unobserved heterogeneity, and accordingly to enhance the overall model performance. Modeling results show that given the same flow rate, the shoulder lane has the highest breakdown probability, and the center lane has a higher breakdown probability than that of the median lane. It is also found that if censored data is ignored then breakdown probabilities would be overestimated. The proposed model can assist diagnosing bottlenecks with frequent semi-congestions which would otherwise be neglected by using the cross-section-based models and can also facilitate the implementation of more effective lane-based traffic control strategies.