An iterative two-stage approach to modeling vacant taxi movements : formulations and implications

Abstract

Taxi services play an important role in transportation systems by providing 24-hour door-to-door services. However, taxis contribute significantly to congestion and air pollution problems, especially when excessive taxis circulate on roads searching for customers. It is thus necessary and important to properly model how taxis navigate on urban road networks in order to establish effective taxi regulation policies to mitigate these problems.

Various taxi network models have been developed to investigate the consequences of a variety of taxi regulation policies. However, very few studies have sought to validate the models based on the real-world data. This thesis uses the global positioning system (GPS) data from 460 Hong Kong urban taxis for calibrations and validations of the proposed search models. At first, the zone choices of vacant taxi drivers in search of customers are modeled to determine the significant factors that influence their search decisions and to demonstrate the variations in their customer-search behavior at different times of day. Then, the sequential zone choices of vacant taxi drivers traveling to adjacent zones of their current locations for customers are further modeled to estimate the search paths taken to reach the customers from the zones visited sequence. Other than revealing the zonal distribution of vacant taxis, a cell-based logit-opportunity model, which incorporates the modeling principles of logit-based search model and intervening opportunity model, is also formulated in this thesis to predict the local cruising movements of vacant taxis, and to estimate the required circulation distance and time to reach a customer.

Based on the contributions of these search models, a two-stage modeling approach is consequently introduced in this thesis to consider both the zonal and local taxi customer-search decisions simultaneously. The first stage problem estimates the zonal distribution of vacant taxis and the second stage problem depicts the local search movements of vacant taxis in a cell-based network within the chosen zone in the first stage. The proposed model formulation is found to offer a great improvement in prediction accuracy on the zonal distribution of vacant taxis to the existing logit-based search models. The two-stage model can also be applied for forecasting the effects of adjusting the fare and fleet size to the vacant taxi trip distribution. Potential taxi policies are investigated and discussed according to the findings, and that provides insights for managing the taxi market of Hong Kong and other international cities with similar urban taxi services provided.