About us

10

NPS Score

200+

Successful AI projects

75+

Scientific publications

Overview

Project scope

We developed an AI-based solution for identifying travel anomalies and predicting stops, for example due to an unexpected delay, diversion or disruption of a train. These disruptions have a negative impact on customer satisfaction. Our solution is intended to support planning in identifying such faults.

Key Objectives & Deliverables

1

Trip data analysis: statistical evaluation past trips and feature engineering.

2

AI-based anomaly detection: Precise detection of driving disruptions, stops and so-called ghost-stops.

3

Interpretable results: Output of detected anomalies supported by specifying probabilities and underlying causes.

Project phases

Our solution

Deutsche Bahn AG (DB) is one of Europe's largest mobility and logistics companies based in Germany. It operates rail passenger and freight transport, connects cities and regions and promotes sustainable mobility. With cutting-edge technology and an extensive route network, DB focuses on efficiency, punctuality and customer satisfaction. Find out more about the Deutsche Bahn.

Mobility & AI

Rail transport benefits greatly from AI-based innovations. According to Deloitte, the market for intelligent mobility will grow to 850 billion euros by 2025, over 1% of global GDP. AI optimises timetables, detects anomalies and disruptions at an early stage, and improves operational processes. It reduces outages, increases energy efficiency and integrates renewable energy — for reliable, sustainable and future-oriented mobility.

Challenge

For the clientOur approachIndustry perspective

The aim is the reliable and timely detection of stops and unusual driving anomalies. So-called ghost trains represent a particular challenge here. These are cancelled or delayed trains which - for example due to missing failure messages - are still displayed in the information systems but never appear on the track. Here, it is important to efficiently support scheduling and to enable discrepancies in passenger information to be resolved quickly and precisely.

"Perelyn supported us in developing an AI solution to detect service disruptions and stop anomalies in long-distance rail traffic. The collaboration was efficient, solution-driven, and technically outstanding. We were especially impressed by the transparent results. Perelyn is a reliable partner for data-driven mobility solutions."

— Sven Krause, Program Manager Passenger Information, Deutsche Bahn AG

Our approach

Together with the Traveler Information Team (DB AG), Perelyn has developed a tailor-made, AI-based solution that analyses all daily long-distance journeys in real time and investigates them for anomalies. This includes current events, such as standstill on a route, as well as historical travel data. Our technology filters relevant information from the flood of data for each train.

Industry perspective

AI-based systems provide the mobility industry with accurate forecasts around the clock, detect anomalies in real time and increase operational efficiency. They reduce operating costs, optimize resources and improve passenger communication. At the same time, they enable valuable data analyses for predictive maintenance. Despite these benefits, data protection and data quality remain key challenges that require careful implementation.

... for a 5 billion railway company, AI could add around 700 million euros in value per year.

Source: McKinsey & Company – The journey toward AI-enabled railway companies

Project phases

Phase 1
Readiness & Prototyping

Preliminary analyses for feature selection and definition of the required range of functions: Statistical analyses to select suitable characteristics for detecting driving anomalies/stop failures.

Development of a minimal prototype: Demonstration of the possibilities of data-driven anomaly detection based on the available features.

Phase 2
Implementation & Deployment

Deploy on AWS EC2: Integration of the containerized solution into the existing AWS Cloud system.

Knowledge transfer: Detailed documentation and interactive code tutorials.

Phase 3
Improvement & rollout

Optimizing the interpretability of results: Illustration of recognition results on probability figures calculated from historical stop failure data, including relevant characteristics.

Phase 1
Vorbereitung & Prototyping

Vor-analysen zur Merkmalsauswahl und notwendigen Funktionsumfangs: Statistische Analysen zur Auswahl geeigneter Merkmal für die Fahrtanomalie-/Haltausfallerkennung.

Entwicklung eines Minimalprototyps: Demonstration der Fähigkeiten.

Phase 2
Implementierung & Bereitstellung

Bereitstellung auf AWS EC2: Integration der containerisierten Lösung in das bestehende System von AWS Cloud.

Wissenstransfer: Detaillierte Dokumentation sowie interaktive Code Tutorials

Phase 3
Verbesserung & Rollout

Optimierung der Ergebnis-Interpretierbarkeit: Erkennungsergebnisse werden abgebildet auf reale Wahrscheinlichkeiten von historischen Haltausfällen unter gegebenen Bedingungen

We find the causes of your anomalies!

Contact us
Contact us

Solution

OverviewBusiness impact & benefitsTechnical architecture & integration

Overview

Automatic anomaly and stop failure detection was developed to efficiently support scheduling and to enable discrepancies in passenger information to be resolved quickly.

Perelyn implemented a full-stack solution together with the Traveler Information Team (DB AG) within their AWS environment, with a particular focus on transparency in the presentation of results, as well as easy configuration and insight by various levels of the organization.

Business impact & benefits

Our AI learns from historical trip data, automatically detects anomalies and predicts possible stop failures in real time. This increases the efficiency of anomaly detection and supports predictive operational management.

Technological approach: Modular & scalable

Our modular, container-based architecture can be flexibly adapted and processes large amounts of data efficiently. A user-friendly configuration enables easy implementation — even for less tech-savvy users.

“Our goal was to make a tangible contribution to supporting dispatchers, to greater customer satisfaction, and therefor to a future-oriented mobility.”

– Michael Banf, Chief AI Scientist Perelyn

Key Tech

Software Engineering

AWS Bedrock, Amazon Comprehend, Google VertexAI, Azure Open AI, HuggingFace Transformers, LangChain, LangSmith, LLamaIndex

Graphdatabases

Pandas, Scikit-Learn, Pytorch, Tensorflow

Machine Learning Operations (MLOps)

Airflow, MLflow

Cloud & Microservices

AWS, EC2, Docker

Contact us

Name*
Email*
message*
Cancel
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Help us improve our website. By “Accept” click, do you agree that we store a cookie on your device to analyse the use of the website. Read our Privacy statement for more information.
SettingsRejectAccept
Something went wrong. Please try again.