Modeling local taxi customer-search movements in a cell-based network

Abstract

OBJECTIVES: Taxis are an important transportation mode but they tend to circulate around a city to search for customers. The circulation activity occupies spare road spaces and worsens traffic congestion and air quality. Numerous studies have been conducted to attempt modeling the route choices of drivers based on taxi GPS data. They can be used to model the local movements of drivers within a zone. However, most of these efforts have been put on modeling the route choices of occupied taxi drivers with a known destination and these studies did not model vacant taxi drivers' route choice behavior for finding the next customers. Without modeling the local customer-search movements of vacant taxis, effective and appropriate taxi regulation policies cannot be established to tackle the traffic congestion and air pollution problems. DATA AND METHODOLOGY: This paper proposes a cell-based model to predict local customer-search movements of vacant taxi drivers, which incorporates the modeling principles of the logit-based search model and the intervening opportunity model. The logit-based modeling concept is used to handle multi-directional, discrete search choices for vacant taxi drivers and the intervening opportunity model is reformulated and incorporated into the logit-based search model to capture the cumulative probability of successfully meeting a customer in each direction of customer-search. The proposed model assumes that vacant taxi drivers do not have clear destinations in customer-search after they drop off their preceding customers, but they travel towards an area with a high probability of meeting a customer. The local customer-search movements were extracted from the global positioning system data of 460 Hong Kong urban taxis and inputted into a cell-based taxi operating network to calibrate the model and validate the modeling concepts. EXPECTED RESULTS: The model results reveal that the taxi drivers' local search decisions are significantly affected by the (cumulative) probability of successfully picking up a customer along the search route, and that the drivers do not search their customers under the random walk principle. The proposed model helps predict the effects of the implementation of the policies in adjusting the taxi fleet size and the changes in passenger demand on the customer-search distance and time of taxi drivers.