Bus Track Restoration Based On IC Card User Travel Mode

The continuous deterioration of urban traffic calls for developed and efficient public transportation systems as a measure for the alleviation of transport issues. The advancement of ICT(Information & Communication Technology), on the other hand, has resulted in a massive volume of GPS trajectories and smart card transaction records and numerous researches on public transport planning and travel characteristics have been inspired in wake of such data flourish. However, due to constraints in hardware and software capabilities, loss and errors in GPS data occur during collection phase. Analytical results would deviate from reality significantly if data incompleteness is not dealt with. Therefore, this thesis targets the solution for an effective recovery of missing bus GPS traces.Through an extensive survey on the bus GPS and smart card datasets, it is found that roughly one-third of the smart card transactions are out of matched GPS traces of buses, which would lead to serious difficulties in follow-up analyses. In this study, data pre-processing is firstly conducted to link previously isolated datasets, including bus GPS traces, geo-locations of bus stops, and smart card transactions of users, together. Then, based on commuting pattern(reoccurring trips between homes and workplaces) and transfer pattern(transfer from a vehicle with traces to a vehicle without traces and vice versa) of users in time and space, data recovery is realized by spatiotemporal constraints extracted from those patterns. Because of the inevitable uncertainty in commuting and transfer pattern, fuzzy set theories are adopted in this study to model the travel patterns of users with analysis of membership functions and degree of membership. Finally, using obtained fuzzy set models, bus GPS traces are recovered and arrival information is acquired through a K-means clustering of the recovered trajectories. Cross-validation is carried out and results show an increase in data availability from 60% to 80% approximately. Data quality is largely improved through the proposed methods and it implies that travel patterns of cardholders are highly useful in recovering bus trajectories. Following data recovery, the OD(Origin-Destination) and transfer mode analysis is conducted using the recovered dataset.This study not only benefits the data quality of bus GPS traces and the follow-up analysis and mining, but also provides new methods to model travel pattern with spatiotemporal uncertainties using fuzzy sets which lays groundwork for future researches.