A Consolidated Knowledge- and Data-Driven Model for Public Transport Service Load Estimation Using Multiple Data Sources

Estimation of passenger flows is the basis for the design, operation, and adjustment of a public transport network. The advancement in sensing technologies enable transit operators to monitor the variabilities in passenger flows continuously and consistently. However, the incomplete, inaccurate, and biased data from automatic sensors pose a challenge to effectively use existing data-driven models to estimate passenger flows. This paper proposes a consolidated knowledge- and data-driven model, which incorporates complex demand-supply interactions in a supervised machine learning procedure. Domain knowledge includes the development of essential demand and supply features from various data sources and the modelling of the relationships between boarding flows, alighting flows and passenger loads on-board. The methodology is applied to three tram lines in Melbourne, Australia, where different types of shortcomings exist in the automated data due to non-card users, free service zone, and possible fare evasions, among others. The test results indicate that the proposed model outperforms the pure data-driven model and is able to be transferred to estimate the passenger loads on a different route. We also identify the key factors that are closely correlated with boarding and alighting flows. The proposed modelling approach would allow transit operators to utilise big data from alternative sources for accurate estimation of service utilisation, especially when single source automated data is not complete and include measurement inaccuracies. The crowd information will not only help operators to accommodate the variability in passenger demand, but also assist passengers in travel planning to avoid service overcrowding