RSSB - Rapid Evaluation and Planning Analysis Infrastructure for Railways (REPAIR)

Project Description

Planning and scheduling of trains are very complex tasks due to their highly specialised and combinatorial nature. In the UK, the required information for planning and scheduling is very fragmented. Controllers must deal with a large number of short-term requests and must consider a myriad of infrastructure constraints, consulting a variety of databases and paper-based documents to complete train planning tasks which increases the likelihood of network delays and is inefficient.

Train planning relies on a number of key steps:
1. Predict the knock on effects of timetable changes (additional freight lines, line incidents, crew shortages etc);
2. Determine an appropriate high level mitigation plan/updated timetable;
3. Development of a detailed timetable plan (if there is time), if not provide routes for each service on a case-by-case basis;
4. Validation that the updated plan can be implemented both by the train crews and into the signalling systems (if there is time).

The proposed solution, REPAIR (Rapid Evaluation and Planning Analysis Infrastructure for Railways), will revolutionise dynamic train planning through the development of rapid predictive analysis, based upon deep learning research and proven Artificial Intelligence (AI) optimisation techniques taken from the Contractors experience in the defence industry, combined with the Subcontractor’s significant railway database. This approach will enable rapid and accurate predictions of the knock-on impact of timetable changes to be generated, and through the applications of a deep Reinforcement Learning (RL) AI approach, potential mitigation options will be suggested. Given an incident, the AI based algorithms will predict delays and its propagation at various timing points in the rail network. This knowledge will empower controllers to make faster and more optimal decisions around mitigation plans or determine specific routes that should be adopted, ultimately leading to faster and better dynamic train planning.