An Analysis Model And Algorithm For Most Profitable Taxi Travel Paths

Taxies are an important means of the public transport in a city and a great supplement to such large-scale urban passenger transportation systems as buses, subways, and light rails. For the taxi industry and the healthy development of the whole city, it is of great significance to develop some scientifically based methods for efficient, real-time taxi routing within a given road network environment. Taxi trajectory data is now increasingly available in the form of GPS records, which is capable of capturing the traffic dynamic of urban road networks. Because of that, they are often used as floating vehicle data in transport analyses, traffic management, and real-time intelligent transportation information services.This paper aims to study the issue of how to detect most profitable taxi routes from historical taxi trajectories. Sample data were collected from four major taxi services in Shanghai, consisting of more than10,000GPS records for one day (24hours) taxi operations in2009. The data were first preprocessed for coordinate conversion, OD matrix extraction, and map matching (based on a self-developed method of combining minimum distance and polynomial curve fitting analyses). Efforts were also made to improve the Dijkstra shortest path analysis algorithm for index computation and statistical analyses in the subsequent route modeling. The modeling process and algorithm development mainly focuses on the following two aspects:(1) Computer simulations were performed to reconstruct historical taxi service spatial dynamics based on the extracted taxi travel OD matrices. The spatial and temporal attributes of taxies were then generated by joining the city road network to the map-matched GPS trajectory data. The24hours of the sampled day were grouped into seven time periods, which allowed temporally sensitive taxi cruising path models to be created for improving the taxi utilization rate and reducing taxi operation costs.(2) A taxi operation econometric model was built by considering the actual operation modes and price tables collected for the study region. Model parameters were rectified with actual cost and traffic information collected by taking taxies to different parts of the city in different time slots. A real-time prediction model for most profitable taxi routing was built by integrating both trip length and profitability factors. As empirical tests, three trips of different time lengths were selected from the historical taxi GPS data to conduct a comparative analysis between model outputs and actual routing decisions made by the taxi drivers.The thesis is organized as follows:The first chapter analyzed the research background and significance and reviewed the related researches from both domestic and abroad sources. The research objectives and content were narrated, and the layout of the thesis was provided.The second chapter introduced the concepts of urban intelligent transportation systems and such related technologies as various vehicle positioning technologies and Geographic Information Systems. The data characteristics and theoretical foundations of common taxi operation systems were explored in details.The third chapter elaborated the data acquisition and preprocessing steps in model building, and introduced an array of related methods, including geodetic coordinate conversion, GPS error model, trajectory-to-road projection and map matching algorithms.The fourth chapter introduced the model implementation workflows, and demonstrated the results of model applications with sample trajectories for both the taxi cruising route model and the most profitable taxi route model.The last chapter summarized the accomplished research work in the thesis and highlighted the innovative points as well as rooms for future improvement.