Sie können Operatoren mit Ihrer Suchanfrage kombinieren, um diese noch präziser einzugrenzen. Klicken Sie auf den Suchoperator, um eine Erklärung seiner Funktionsweise anzuzeigen.
Findet Dokumente, in denen beide Begriffe in beliebiger Reihenfolge innerhalb von maximal n Worten zueinander stehen. Empfehlung: Wählen Sie zwischen 15 und 30 als maximale Wortanzahl (z.B. NEAR(hybrid, antrieb, 20)).
Findet Dokumente, in denen der Begriff in Wortvarianten vorkommt, wobei diese VOR, HINTER oder VOR und HINTER dem Suchbegriff anschließen können (z.B., leichtbau*, *leichtbau, *leichtbau*).
In diesem Kapitel wird Allié Supervision vorgestellt, ein Instrument zur Entscheidungsunterstützung, das Eisenbahnunternehmen bei der Bewältigung von Störungen und der Minimierung von Verspätungen unterstützen soll. Das Tool nutzt einen makroskopischen Simulator, um Zugbewegungen vorherzusagen und die Entscheidungsfindung in Echtzeit zu unterstützen. Der Text diskutiert den Stand der Technik in Eisenbahnsimulatoren und hebt die Zielkonflikte zwischen Genauigkeit und Rechenzeiten hervor. Darin wird das Simulationsmodell von Allié Supervision detailliert beschrieben, das Merkmale wie Beschränkungen des Fortschritts und Verzögerungsmechanismen enthält. Die entwickelte Lösung stellt eine benutzerfreundliche Anwendung dar, die es Betreibern ermöglicht, mit der Simulation zu interagieren, zukünftige Störungen einzugeben und verschiedene Szenarien zu vergleichen. Die Experimentierphase umfasste reale Tests in einem operativen Zentrum, wo das Werkzeug aufgrund seiner Fähigkeit, Verzögerungen und potenzielle Konflikte zu antizipieren, gut ankam. Die Schlussfolgerung skizziert Zukunftsperspektiven, einschließlich der Integration des Simulators in ein simulationsbasiertes Optimierungsrahmen und der Erforschung stochastischer Simulationen für verbesserte Verzögerungsschätzungen.
KI-Generiert
Diese Zusammenfassung des Fachinhalts wurde mit Hilfe von KI generiert.
Abstract
SNCF Voyageurs/TGV-Intercités operates over 800 highspeed trains (TGVs) per day in France. The system can be highly tense during peaks with up to 13 trains per hour on the same track section. Thus, minor delays can affect operations, especially for long-distance trains. Supervision of the system and real-time rescheduling are difficult tasks given the complexity of the rail network and the cohabitation of different train services.
In this work, we present a real-time decision-support tool providing estimations on the arrival time of trains at each station and at destination and helping comparisons between rescheduling choices.
The estimations of future delays have a focus on explainability, with information on the causes of the delay and on the possibility to recover the delay. The operators can use the tool to test different rescheduling choices and see the impact of each scenario on the traffic, helping them deciding the actions to take to minimize delays. The predictions of arrival times of trains are based on a macroscopic discrete event simulator, coupling the theoretical timetabling with real-time information on the trains’ positions.
Arrival times are displayed on the interface of the developed web application, where the user can interact with the simulation by adding information and comparing different disruption management scenarios.
The tool has been tested with success in an operational environment. The operators gave positive feedback on the tool, underlying its capacity to give them more insight on the expected delay evaluations and on potential conflicts. The information displayed was judged relevant and reliable. This is confirmed by our analysis of the quality of the simulation.
1 Introduction
Within SNCF Voyageurs, the main French passenger rail company, TGV-Intercités operates every day 800 long-distance trains in France and Europe under the brands TGV Inoui and OuiGo. Five operation centers assure the supervision of trains and staff. The operators take actions in case of disruptions on the train lines in order to minimize delay propagation, ensure the resource management, inform the passengers and organize their take-over if their journey is disrupted.
Operators can take several actions to reduce the consequences of a perturbation: cancelling a train, changing its path to an alternative route, changing the scheduled order at a junction, changing a scheduled coupling. These actions are taken with specific goals in mind in terms of minimizing passengers delays and maximizing the efficiency of resource management.
Anzeige
When a disruption occurs, operators have a high cognitive load and must take complex decisions in a short time. Today, they lack the tools to predict the future train circulation and plan the operations, which would help them in the decision-making process.
In this paper, we present a decision-support tool, named Allié Supervision, which is based on a railway simulator and aims at giving the operator support in the supervision and the decision-making process.
The paper is organized as follows: Sect. 2 presents a brief state of the art of railway simulators for real-time applications; Sect. 3 describes the main features of the decision-help application that we propose to the operators; Sect. 4 details the experimentation we conducted in an operational center, where the tool was given to operators for real-life tests, and Sect. 5 provides conclusions and perspectives.
2 State of the Art
Many railway simulators exist in literature, both commercial or open source. An extensive review of railway simulators can be found in [3].
Anzeige
Microscopic models, like OpenTrack [1] or Denfert [2], propose a detailed modelling of the infrastructure and of the trains circulations. These models provide accurate simulations, on condition that all the necessary parameters are available with sufficient precision. The accuracy comes at the cost of higher computations times, which makes these kind of simulator unsuitable for a real-time application, like the one we are presenting in this paper.
In comparison, macroscopic simulators offer faster computation times, at the expenses of a loss in certain details of the infrastructure or of the train path. An example of such simulators is Prism [6], which provides a Monte-Carlo approach, and [5], which provide an accurate comparison between a macroscopic and a microscopic approach.
Machine-learning based models also exist, such as Transformers [4] which uses the homonyms deep-learning architecture.
Our approach is novel in which it provides a focus on real-time applications in case of disruptions, giving the operators the ability to interact with the simulator and compare different operation scenarios. It also focuses on the explainability of the forecasts.
3 Allié Supervision
3.1 Simulation Model
The main feature of Allié Supervision is its use of a deterministic, macroscopic discrete-event railway simulator to forecast current delays and actions injected by the users. We used a simulator developed internally, whose main advantages are as follows:
its convenience when deploying on new perimeters
its mechanism to monitor the causality between delays,
its modular design in regards to simulation rules.
In Allié Supervision, the simulation rules include headway constraints, delay catch-up on running and stopping times, and a heuristic to schedule delayed trains at junctions and on station departure. These rules are mostly parameterized and are calibrated ahead of time in order to maximize the forecasting accuracy of the simulation. These rules can either prepone or postpone the planned time of events, reducing or increasing simulated delay respectively.
In our model, the simulated events are departures, passings and arrivals of all simulated trains. In order to reduce computation times, these simulated trains are usually a subset of all the trains running daily in France. This subset is parameterized to include most trains running in the geographical area of interest.
At each run of the simulation, the first simulated event of each train is the one matching its latest observation. Trains that have not yet departed or been observed are initialized at their originating station and are supposed to leave on time. For all these trains, all successive events are simulated. Hence, in a real-time context, we simulate all events that will occur between the current time and the end of the operating day.
From the simulation are retrieved simulated times for each event and the causes of variation delay between consecutive events.
3.2 Developed Solution
A tailored application has been developed to present the simulation results to the operators, and to allow them to interact with the simulation in a simple way. The ergonomics of the application are crucial in a real-time context, where the cognitive load of operators is high and increases in case of disruptions.
The software architecture is presented in Fig. 1. The entry data are the theoretical timetable and the real-time positions of trains. A new simulation is launched every time a train position update is received. The simulation outputs are used to present to the operator with relevant information on the traffic:
Global KPIs on the number and types of delayed trains for different time ranges are shown to give the operator an overview of the current and forecast system status
A detailed view of each train circulation presents the estimated arrival times at each stop and includes explanation on the causes of variation of the delays: for example, the delay of a train could be reduced by using the running time supplement, or could increase as a consequence of other trains circulations (scheduling at junctions, spacing)
The operator can interact with the simulation by adding information on a future disruption, such as a delay on a train or on a train path or the rescheduling at a junction. The simulator will integrate this new information and allow the operator to see the consequences on the circulations and to compare different scenarios. A notification function is also available to alert operators when needed, thus minimizing additional cognitive charge.
4 Experiment Phases
4.1 Experiment Perimeter
Experiments have been recently conducted with the COS Sud-Est, the operating center monitoring TGV operations in the south-eastern part of France. The simulation perimeter includes other high-speed trains running in that region (Ouigo, Trenitalia), intercity trains, regional trains and freight trains. These trains add up to about 2500 trains daily, among which about 270 are supervised at the COS.
4.2 Simulation Performance
Figure 2 presents the distribution of running times for the simulator depending on different time-ranges across the day. It can be seen that computation times are in the order of a second. They decrease throughout the day because we do not simulate events preceding the latest observation for each train.
Fig. 2.
Computation times of Allié Supervision on the COS SE perimeter by specific hour ranges
Ahead of user experiments, we also measured the accuracy of our simulations to propagate delay hypotheses injected by the users. We used one week of past observation and timetables to extract past incidents to base our simulations on. We extracted from this dataset occurrences of delay increases higher than 5 min, that we interpreted as primary delays. Each of these delays was injected in the simulation after which were retrieved simulated arrival times at all the following stops of the impacted train. From 400 primary delays collected, we obtained a set of 1000 simulated arrivals.
We compared our predictions to “translation”, a dummy predictor that predicts that the delay of a given train to its future stop will be its current delay. The metrics that are used are
MAE: mean absolute error
Fiability: rate of predictions whose errors are below 5 min (indicator used in operations to evaluate passenger information accuracy)
Table 1 presents an evaluation of the predictions of our simulator to this set of primary delays. The fiability of TGV operators falls between 60% and 65%. However, the actual score cannot be directly compared to our results since they were obtained on a different time period, but we expect Allié Supervision to improve it.
Table 1.
Prediction accuracy of our simulation compared to “constant prediction”
Metric
Our simulation
Translation
MAE
\(324^{\prime \prime }\)
\(461^{\prime \prime }\)
Fiabilité
71,3%
57,3%
4.3 Results
Six weeks of experiments took place at the COS Sud-Est to evaluate the relevance of our solution and retrieve user insights from the operators. The users appreciated our solution in addition to their current environment, noting that Allié Supervision helped them to better anticipate delays propagation and future potential conflicts. Longer experiments would be required to measure how much our solution improves the operators’ performance.
The possibility to inject information in the simulation was less tested by the operators, but was found helpful to sharpen the delay estimations in specific situations, for instance when maintaining connections with delayed trains.
Initially developed as an aid to passenger information, the tool revealed other uses when put in the hands of the operators. Usually, when a delayed train is bound to attach a train on time, the operators can cancel the attach operation to save the train on time. However, on such an occasion, Allie Supervision predicted that a train with 20 min of delay could reduce it to 10 by the coupling station. The coupling was maintained and the two trains arrived on time, allowing a better use of resources and hence a financial gain.
5 Conclusion and Perspectives
In this paper, we presented a novel solution for monitoring railway traffic based on real-time macroscopic simulation. Our proof of concept was tested in TGV operating centers and validated by our users.
Our interpretation of simulation outputs as explained delay forecasts yielded good accuracy, better than the simplistic rule currently used, and facilitated the acceptation of the predictions by the users.
The functionality to inject hypotheses to the simulation was positively perceived by the users but noted that the UX could be improved. We also plan to integrate and test new types of hypotheses, such as rerouting and train scheduling on double-way tracks.
Future perspectives are to include the simulator in a simulation-based optimization framework to add recommendation functions to our tool, and explore stochastic simulation to strengthen its delay estimations.
Acknowledgements
This project was co-financed by SNCF DTIPG and TGV-Intercités. The authors thank the TGV agents at COS Est and COS Sud-Est for their time and constructive feedback on the prototype.
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.
The images or other third party material in this chapter are included in the chapter's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the chapter's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
Huerlimann, D., Nash, A.: Railroad simulation using OpenTrack. WIT Trans. Built Environ. 74, 10 (2004)
2.
anticiper la robustesse des plans de transport: C. lerin c. leboeuf. denfert. Revue générale des chemins de fer 294, 26–37 (2019)
3.
Ho, T.K., Goodman, C.J., Siu., L.K.: A review of simulation models for railway systems (1998)
4.
Pierre, A., Arthaud, F., Lecoeur, G.: Transformers à grande vitesse (2021)
5.
Warg, J., Solinen, E., Bohlin, M., Zinser, M., Betz, T.: Comparison of microscopic and macroscopic approaches to simulating the effects of infrastructure disruptions on railway networks (2018)
6.
Geilke, M., Zinser, M., Betz, T.: PRISM: a macroscopic monte carlo railway simulation. In: Proceedings of The 12th World Congress on Railway Research (WCRR) (2019)
