Project Summary
The FAST project aims to supply accurate, real-time carriage information on how busy a train is, using historical and live weight data to provide capacity estimates on rail, enabling passengers to choose whether to travel and, if so, where to board.
Project Achievements
-
Designed, implemented and optimised the configuration of a Long Short Term Memory (LSTM) Neural Net, which takes historic loading data, real-time loading data from up-line stations and contextual information e.g. weather as input. Detailed historic loading data was provided by train operating company Southeastern.
-
Implemented a Last Values predictor, which uses solely historic data, to provide a benchmark for assessing the performance of the Neural Net. Assessed accuracy of all predictors using Coefficient of Determination (R2).
-
Found that including up-line loadings significantly improved the Neural Net predictions and that the Neural Net predictor consistently outperformed the Last Values predictor, with or without real-time up-line loadings.
Next Steps
-
Adopt a test-driven approach to enable the team to increase the reliability and scalability of the technology, readying it for commercial application. The technology will be able to help operators manage services in the event of network disruption.
-
Promote the technology to all UK operators interested in improving crowding information for passengers, and liaise with the Department for International Trade on potential exploitation overseas.
