Link-Data-Based Approximation of Path Travel Time Distribution with Gaussian Copula Estimated Through Lasso

In this paper, we highlight the characteristics of floating car data, and extend our research on single link to pathes. By the law of total probability, we proposes to approximate the total path travel time distribution by the probability-weighted sum of a series of conditional path travel time distributions, conditioning on the sequence of entering time to the links of the path which is expressed as its corresponding (entering-time) lag vector.Any of such conditional distribution is the sum distribution of a sequence of link travel time distribution with specific dependence structure. The dependent structure is modeled by a lagged Gaussian copula while the marginal distributions are estimated by kernel method.The L1-constrain-minimization (Lasso) method is utilized to obtain an invertible covariance matrix of the Gaussian copula even for limited data. Compared to the iterative type procedure, this approach is efficient when the number of scenarios to visit is limited and it resolves the "conditional distribution puzzle" in the iterative formulas