A Maximum Likelihood Method with Penalty to Estimate Link Travel Time Based on Trip Itinerary Data

Travel time is an important network performance measure. It is a challenging subject due to the fluctuations in traffic characteristics, such as traffic flow. This study proposes a maximum likelihood method with penalty to estimate link travel time based on trip itinerary data from a statistical point. Three penalized models, which are Lasso penalized model, Ridge penalized model and Revised-Lasso penalized model, are introduced. The models are discussed and compared with the basic model which is a maximum likelihood function without penalty. First, the predictive performance of the basic model and three penalized models are evaluated based on the data of three simulated networks. Results suggest that Revised-Lasso penalized model outperforms other models. In this research, Revised-Lasso penalized model is applied to a simplified Sioux Falls network. This study also provides a detailed procedure to estimate link travel time parameters in the simplified Sioux Falls network. Finally, the effect of the sample size on estimation accuracy is tested. The results show that sample size has a significant effect on the basic model estimation, but it has little effect on the Revised-Lasso penalized model estimation. This study provides an efficient and accurate way to estimate link travel time distribution.