Simulating Disruptions on Netherlands Railway Operations

Simulating Disruptions on Netherlands Railway Operations

Overview

Netherlands Railways are one of the busiest rail networks in the world after Switzerland. Their revenue for 2018 was 5.926 million euros. Every year, 9 million people travel on blue and yellow trains. The railways run ‘from Utrecht to Tibet’ and spread out for 6,830 kilometers.

Problem

Netherlands Railways engineers used virtual testing of innovations. They could not try out many innovations in real life. Also, they could not cause disruptions on railway lines because it would destroy the entire country. Netherlands Railways engineers needed to test the procurement and introduction of the new rolling stock and innovations such as self-driving trains and the digital train protection system.

Netherlands Railways previously used only microscopic, very detailed simulations, but it was hard to work with them.

Disruptions were one of the main problems for Netherlands Railways. They are hard to predict, but happen every day. Every day, there were restrictions. There was no day when all trains were on time with the right crew and the right rolling stock. Disruptions could be cows on the track, doors not closing, or power failures. For example, a high-voltage cable dropped on the railway track, and it took half a year to repair it. Usually, they simplified problems into the following: a broken train blocked the track, or the track was blocked and no trains were allowed to pass.

Solution

Using AnyLogic, they made macroscopic, conceptual models. Engineers could develop a concept rapidly, spending less than one week on a new model. They integrated AnyLogic with a lot of other predictive tools. The model was interactive, easy to use, and explainable to the end users.

The AnyLogic simulation model was much better than the visualization of the railway network that was previously used by Netherlands Railways. The model simulated all the trains for the whole day.

They modeled one line from the network that they were interested in. Then they load a timetable from any year, past or future, e.g., 2030. They simplified this data into bits of infrastructure, stations, platforms, and trains.

The model developers loaded real data from 2019 and fitted it into stochastic distributions. The AnyLogic model was used to predict the probability that a train would be on time and simulate the flow of trains.

Schematic track layout

Schematic track layout

Netherlands Railways developers built a simulation model to predict the network-wide punctuality that could not be done by hand. The AnyLogic model provided realistic breakdown rates for trains and infrastructure. Netherlands Railways developers used some simple rescheduling rules.

Macroscopic railway layout

Macroscopic railway layout (click to enlarge)

In case of an incident, passengers who were not in the affected zone could continue to be transported. It was important to limit the disruption to as small an area as possible and to turn trains at the edges to keep passengers on time.

Moreover, Netherlands Railways developers ran a Monte Carlo experiment. In the model, there were 10 stochastic distributions. Every time the train was departing, it showed possible delays or failures. It was connected to the SQL database to load parameter sets that were predefined.

Results

The engineers were able to test certain kinds of disruptions. When an experiment was run, the developers got the KPIs, that were the total delays of the trains, the number of failures, cancelled trains, and punctuality. The running time of the model was quite good, as this experiment showed the results for 5 years in around 5 minutes of running it.

Monte Carlo experiment with AnyLogic

Monte Carlo experiment with AnyLogic (click to enlarge)

Netherlands Railways has been using the model for almost 5 years. During this time, they made some changes to it. For instance, they used actions more in the transition instead of the state entry. It prevented confusion when using historical states.

Previously, Netherlands Railways engineers drew the model layout by hand, gave it a name, and connected it to an input file, but then they started to make automatization with the use of AnyLogic templates that enabled them to switch easily between layouts (parts of the Netherlands). Thanks to AnyLogic, it became possible to model almost the whole of the Netherlands.

To get inputs from the timetable planning software and the rolling stock scheduler, the model was connected to some of their production systems. Netherlands Railways used XML files, converted them to a SQL database, and then AnyLogic got the input data from SQL. After the simulation, the output data could be exported from AnyLogic to a SQL database and visualized with Power BI.

Case study: Increasing train frequency

Problem

Furthermore, Netherlands Railways were asked by the government to increase the frequency of the trains while maintaining the same KPI, which was punctuality level. A busy network always decreased punctuality because there were more trains that could fail, more passengers that could cause delays, and the infrastructure was more heavily utilized.

Solution

Netherlands Railways needed to investigate how to maintain the same high level of punctuality. For this purpose, they used a very abstract model. It included a number of slow and fast trains, tracks, stations, and the real timetable. They simulated different timetables.

Results

The picture below illustrates the percentage of punctual trains that show a performance drop. For example, the orange bar shows four fast trains per hour without building the new tracks. If they added two tracks to the infrastructure, the level of punctuality would be higher.

Percentage of punctual trains

Percentage of punctual trains

One of their challenges was how to drive more trains with the same infrastructure. This performance drop could be compensated not only by infrastructure but also by train reliability and timetable adaptability.

Case study: Train procurement

Problem

Netherlands Railways wanted to buy the double-deck rolling stock the new motor units that required extra investment. The stock was the most capital-intensive asset, so they needed to find out if this investment in reliability was worth the money.

Solution

They simulated a new timetable for 2040 where the frequency would be doubled – around 8 or 10 trains on a track per hour. The model considered the mixed environment with all train types, failure data from 2019, and failure rates of the new trains.

Results

Trains arrival times were one of the KPIs. The customer benefit was punctuality.

The difference in train punctuality for reliability of new trains

The difference in train punctuality for reliability of new trains

If the reliability of new trains was improved by 20%, only 0,08% of modeled trains would arrive on time. The AnyLogic simulation model showed that it was not cost-effective. Thus, Netherlands Railways decided to invest in better incident management.

Other studies

In addition, the engineers conducted a railway optimization study for repair locations. Among the results that Netherlands Railways achieved with the use of AnyLogic as a railway simulation software were the introduction of new rolling stock, the new safety system, automated railway operations, long-term infrastructure investment, and short-term operational decisions.

The case study was presented by Camiel Simons, of Netherlands Railways, at the AnyLogic Conference 2022.

The slides are available as a PDF.


Похожие проекты

Другие истории успеха

Сборник историй успеха от пользователей AnyLogic

Скачать