Project overview
This project built on the earlier RRUKA-funded work looking at the use of passenger loading data to influence behaviour and mitigate crowding. It was funded as part of Future Rail’s TOC’15 initiative.
On the basis that providing better crowding data to passengers seems desirable, the first aim was to identify the most suitable data sources for counting passenger numbers. A range of data sources were obtained, which vary in terms of accuracy, cost and other practical limitations (for example, not all sources can be read in real-time), and work was undertaken to assess their relative merits. The optimal solution may be found by using some combination of the available data sources. The second aim was to develop the methods used to accurately predict passenger loadings from the data available, developing algorithms and working with GTR’s appointed developer to ensure that they can be implemented. The final aim was to determine how best crowding information should be presented to the travelling public and to implement a small-scale trial in order to further assess the benefits.
A set of manual counts were commissioned by GTR in order to be able to calibrate a set of models. Regression models have been developed to estimate the number of people currently on a train from a number of factors, including the recorded weight of the carriages. Software has been written in Python and initial models were developed which predict passenger loadings in real-time and offer forecasts for the month ahead.
On the basis that providing better crowding data to passengers seems desirable, the first aim was to identify the most suitable data sources for counting passenger numbers. A range of data sources were obtained, which vary in terms of accuracy, cost and other practical limitations (for example, not all sources can be read in real-time), and work was undertaken to assess their relative merits. The optimal solution may be found by using some combination of the available data sources. The second aim was to develop the methods used to accurately predict passenger loadings from the data available, developing algorithms and working with GTR’s appointed developer to ensure that they can be implemented. The final aim was to determine how best crowding information should be presented to the travelling public and to implement a small-scale trial in order to further assess the benefits.
A set of manual counts were commissioned by GTR in order to be able to calibrate a set of models. Regression models have been developed to estimate the number of people currently on a train from a number of factors, including the recorded weight of the carriages. Software has been written in Python and initial models were developed which predict passenger loadings in real-time and offer forecasts for the month ahead.
