A New Service Time Framework for Expressway Toll Lane Capacity Model

In expressway systems, critical challenges in toll plaza management arise due to traffic congestion at toll plazas during rush hours. Existing studies often rely on prediction models or simulations, where their accuracy to real traffic conditions is compromised. This study proposes a new framework for the service time to improve the overall analysis of lane capacity calculation for toll plaza management. The investigation uses real transaction data from the three expressways with the highest traffic volume from the Expressway Authority of Thailand (EXAT). The preprocessing includes data cleaning and outlier removal, where the interquartile range method is used to eliminate data that are above the range of realistic service time. Afterward, the upper bound threshold and k-nearest neighbors algorithm are used to learn a traffic pattern model to categorize peak and off-peak traffic. Our new framework is investigated both with real transaction data and in simulations. As a result, the lognormal distribution is best suited for describing the service time. It can improve the accuracy of the service time by 80.59% on weekdays and by 72.64% on weekends for the calculation of the lane capacity. This marks a significant improvement in toll plaza management.