Bus Passenger Origin-Destination Flow Estimation Using Entry-Only Smartcard Data: A Self-Supervised Learning Method Without Alighting Data

To find the passenger Origin-Destination (OD) flow with entry-only smartcard transaction data, the existing rule-based methods estimate the alighting and transfer location approximately, but the accuracy of these rule-based methods may be compromised. To increase the accuracy of the estimation, this paper proposes a learning-based method to estimate real-time OD flow on a bus network. This model is based on the Generative Adversarial Networks self-supervised learning framework, which can allow to learn high-accuracy estimation without using observed passenger alighting data. To show the effectiveness and efficiency of the proposed method, a case study is conducted on the real-world datasets in Tianjin, China, Brisbane, Australia, and Panjin, China. Using the entry-only data in the Tianjin case, results show the effectiveness to estimate accurate OD flow without alighting data. The sensitivity analysis on the proportion of data loss shows that when the loss rate is less than 40%, the proposed method is still capable to provide accurate outputs. Using the ground-truth OD flow data in the Brisbane case, results show the ability to obtain the realistic estimation from flawed OD flow. The robustness analysis results also show the sufficient accuracy for bus planning and operation under various data flaw levels. Using the onboard survey data in Panjin case, the proposed method’s capability of repairing the rule-based estimation is further verified.