The solution to the problem lies in matching Train Services to Rolling Stock.
Train Operators use the term Train Service to refer to a sequence of locations and times at which a particular service is expected to arrive at or depart from. For instance a service A might depart at 7 am from King’s Cross, arrive at 7.10 am at Holloway and depart at 7.12 from Holloway to Finsbury Park and so on. However, it does not state which physical train (or Rolling Stock) will run service A.
Usually, the Train Operators will create a Diagram that states which particular Rolling Stock will run (part of) a service. Under normal circumstances, the operators follow these plans. But when any disruption occurs, everything needs to be re-planned. Depending on the severity of the problem and the size of the company, it is not always possible to keep track of all the decisions that were made.
If no accurate record was kept, at the end of the day the operators need to manually figure out which Rolling Stock ended up running which Service. This is a very time-consuming and error-prone task. However, since Rolling Stock usually have GPS and we know exactly what happened to each Service, we can use that data, together with the Diagrams and the Schedule, to automate this process. A potential algorithm would consume various real-time data feeds and give as an output which Rolling Stock actually ran which Service.
Atos is a European IT services company that provides consulting, managed services and system integration. The company also provides management services to Network Rail. Essentially, Atos provides all of the resource planning, diagramming and the services to distribute data amongst multiple systems. But since the railway is very old and large organization, managing it is a very difficult task. Therefore our challenge is to an essential part of the entire system more efficient and effective.
The two main deliverables in the project can be summarised concisely as:
However, this can be broken down into many more detailed requirements&emdash;these are listed in the MoSCoW format below:
Our prototype as of April 2016 satisfies all of the must and should have requirements. We have not been able to test the system using real-time data as we don’t have access to ATOS systems, but we have tested it on different static data sets. This is as good as it helped us to quantify how well our algorithm works on different datasets.