Modeling lane-level travel time reliability with uncertainty quantification

Identifying Travel Time Reliability (TTR) patterns is vital for analyzing delay probability and traffic uncertainty at the lane level of freeways. This study uses a Bayesian Model Averaging (BMA) copula to model lane-level Travel Time (TT) dependencies, addressing limitations of convolution models. Using detector data from a congested freeway, we evaluate posterior probabilities in a BMA copula model integrating Gamma, Weibull, Normal, Lognormal, and Log-Logistic distributions to capture TTR dependencies. Results show high correlations and Kendall’s tau values between adjacent lanes within a segment, with inter-segment TT dependencies decreasing with distance. The fifth lane, farthest from the curb, exhibits distinct TTR characteristics due to fewer traffic maneuvers. The BMA Student-t copula outperforms the BMA Gaussian copula and convolution model. Posterior weights favor Log-Logistic and Lognormal distributions, reflecting the skewed, heavy-tailed nature of TT data. This approach advances TTR modeling by resolving lane-scale stochastic dependencies and quantifying TT uncertainty.