Improving the estimation of platform wait time at the London Underground

Abstract

In recent years, the proliferation of automatically collected data has allowed many transit agencies to complete more frequent and thorough analyses of service quality. However, while the types and quality of automatically collected data are sure to improve in the future, many transit agencies will be limited to using their current automatically collected data until they have the time and resources to implement new data collection systems. This thesis focuses on improving the analyses undertaken with the currently available data. The primary objective of this thesis is to improve the accuracy of the estimation of platform wait time (PWT) at the London Underground (LU) by determining the methodology that provides the most accurate and robust estimates of PWT. Three methodologies are tested: (1) LU's current PWT methodology using train tracking data that has been made more complete and robust through the use of automated processes; (2) a variant of LU's current PWT methodology; and (3) an improved PWT methodology that avoids the deficiencies of LU's train tracking data. Specifically, this improved PWT methodology relies on the count of trains recorded at stations in order to eliminate the need to use train identification data to verify that a specific train reached a specific destination station and to minimize the effect of data recording errors on the estimation of PWT. The PWT methodologies presented in this thesis are applied to a four-week period on the Bakerloo and Piccadilly lines. For a specific time period and day, it is found that the differences between the PWT estimates from a new PWT methodology and LU's PWT methodology are usually less than 5%. It is concluded that higher quality NetMIS data and improved PWT estimation methods are a worthwhile investment, even if they lead to small changes in estimated PWT, because they ensure that variations in PWT reflect actual operations and are not due to poor NetMIS data or PWT estimation errors. Further, a hybrid approach that combines the best of LU's current PWT methodology and the train-count-based PWT methodology is recommended as one way to improve PWT estimates.