A MAS architecture for dynamic, realtime rescheduling and learning applied to railway transportation

A simulation-based optimization approach to reschedule train traffic in uncertain conditions during disruptions

“…The simplest decisions are based on the planned timetable order, or static priorities (di erentiating between classes of train services); however, in general, better decisions are made based on actual data of the trains, real-time information of the disturbance, as well as operational constraints. Furthermore, the exact time and location of the disturbances may not be known in advance [3]. These facts bring many di culties in designing train dispatching actions or policies.…”

“…Hassannayebi and Zegordi [30] proposed variable and adaptive local search algorithms to minimize the total and maximum waiting time of the passengers for urban rail transit systems. Narayanaswami and Rangaraj [3] designed a multi-agent system model with a learning mechanism for real-time train rescheduling in a bi-directional railway tra c on a single-track route. The developed framework employs a dynamic scheme of priority assignment procedure that allows for dynamically dispatching the disturbed trains in real-time and constructs a deadlock-free disposition schedule.…”

Traffic Conflict Detection and Resolution (CDR) in ERTMS System

“…Globally, the supervisor agent is active all the time on both layers. Narayanaswami and Rangaraj (2015) also evaluated the agent approach by comparing the experimental results with CPLEX solver. It concluded that CPLEX solver can work quite fast when the problem size is small however it requires a very large computational time in some (more difficult) problem instances; whereas all problem instances can be solved in almost uniform time via the multi-agent approach.…”

“…Two main multi-agent architectures (autonomous and mediator architectures) for dynamic scheduling were concluded to simulate behaviour of different agent roles. Concerning the real-time agent-based solutions for railway rescheduling and routing, the authors in (Narayanaswami and Rangaraj, 2015;Proença and Oliveira, 2004) proposed the two architectures of multi-agent system coincidentally with two main layers. In (Proença and Oliveira, 2004) a MAS architecture with a control subsystem aiming at providing secure and efficient routing for all trains and preventing conflicting situations and another sub-system containing all learning and supervisor agents was presented.…”

“…In (Proença and Oliveira, 2004) a MAS architecture with a control subsystem aiming at providing secure and efficient routing for all trains and preventing conflicting situations and another sub-system containing all learning and supervisor agents was presented. In (Narayanaswami and Rangaraj, 2015), the so-called regulating (or rescheduling) layer is incorporated by supervisor, train, station and a new type of agent: auctioneer agent, where an auctioneer agent identifies the conflicting trains for a particular disruption and remains active till that disruption is resolved; the learning layer has also been extended with learner and facilitator agents, where the facilitator agent provides the selected rule to the system after learning. Globally, the supervisor agent is active all the time on both layers.…”