Leveraging data from a smart card automatic fare collection system for public transit planning.

摘要

This research is based on a set of validations data from a smart card automatic fare collection (AFC) system. The goal of the research is to develop new methods in data processing, data enrichment and data analysis in order to better quantify transit demand, enhance operations planning, improve system management and understand travel behaviour.;When studied as an analogue of the regional origin-destination survey, it is demonstrated that smart card data answer more adequately the needs of transit planning in terms of timeliness, coverage and resolution. Other benefits include absence of non-response and respondent fatigue; absence of transcription error; systematic and uniform coding; more precise values and integration of operations data. These properties allow them to be used as a versatile multi-purpose transit survey. Similar to other passive data, certain dimensions of travel cannot be captured. Data processing and enrichment procedures are therefore required.;The dataset contains 763,570 validation transactions from 21,813 cards. A validation strategy is proposed to improve data quality by assuring their internal coherence. This involves error detection and data correction by imputation. The rationale of this approach is to re-establish spatial-temporal continuity, and to avoid the propagation and amplification of error. The error-detection strategy is based on spatial-temporal and public transit logics. About 15% of transactions contain erroneous or suspect values. Run and stop values are corrected by imputation based on the concepts of repetition of scheduled service and the boarding history of individual cardholders. After the procedure, 98.1% of transactions are considered valid as opposed to 84.3% before the procedure.;Several data enrichment steps are undertaken: alighting stop estimation for each boarding with the concept of "boarding chain"; interpolation of stop arrival time for each vehicle run according to temporal information embedded in the transactions and transfer identification with the concept of "spatial-temporal coincidence". These enrichments allow the reconstruction of complete itineraries from boarding data. Other objects, such as activity duration, distance traveled and average speed of a trip, are derived thanks to the totally disaggregated approach. Transfer analysis shows that on a typical weekday, the number of transfers revealed by the AFC system is about 40% higher than those estimated with the concept of "spatial-temporal coincidence".;The association between trip generators and stops are achieved by the multi-day informational approach and spatial-temporal consolidation of itinerary. Multi-day informational approach aims to characterize or interpret trip attributes with respect to all trips within the analysis period. A hotspot analysis reveals anchor points of a cardholder. With this approach, 43% of cards with student fare are assigned to educational establishments. Residence areas are also derived. Each trip is interpreted with respect to a personalized list of anchors. A trip table and a monthly activity schedule can be reconstructed with a lot of details for each cardholder. It also allows travel behaviour comparison between a subgroup of cardholders sharing a common anchor. The multi-day trip characterization leads researchers to rethink some fundamental aspects of trip description. Applications of two data mining techniques, association rules and classification, are proposed for travel behaviour analysis.;Data from the smart card AFC system of the STO are relatively simple and the primary focus of the research is on the validation data. To address this issue, data from a multi-operator and multi-modal AFC system in the Greater Montreal Area, OPUS, are used to illustrate the complexity and technical challenges. They are also used to introduce the potential of other types of data, namely sales and verification data, that are suitable for transit planning, operations and management. Since the setup of each smart card AFC system, the transit network and its fare structure is unique, the needs on data processing and enrichment vary and are specific to each system. However, the principles, the methodological approaches and the analyses proposed in this research can be adapted and transferred to datasets with similar structure. (Abstract shortened by UMI.)