Rail-mind

lastmod 2026-10-21created 2026-10-154 edits

Railway Artificial Intelligence & Logistics Monitoring INtelligent Dispatcher - an AI brain that perceives and judges the entire railway network on its own

Idea

The core solutions for applying AI to the railway authority's train traffic control and passenger behavior analysis fall into two main axes: operational efficiency and safety control (Dispatch & Operations), and in-station passenger movement and behavior analysis (Passenger Analytics).

AI-based train traffic control and operation system

• Real-time conflict resolution and dynamic rerouting (Dynamic Conflict Resolution): When unexpected events such as train delays, track faults, or bad weather occur, a reinforcement learning (RL) algorithm computes thousands of detour routes and priorities. It recommends to the dispatcher, within one second, the rearrangement scenario with the fewest bottlenecks that minimizes total delay.
• Demand-responsive flexible dispatch (Demand-Responsive Dispatch): Instead of relying on a fixed timetable, it links passenger demand forecasts by day of week, weather, and local events with current train positions to adjust headways in real time. It proactively puts spare trains into sections where congestion is forecast to become severe.
• Energy-optimized driving profiling (Energy-Optimized Driving Profiles): AI comprehensively analyzes terrain gradient, curved sections, train weight, and spacing to the trains ahead and behind to derive optimal acceleration and deceleration curves. These are linked with automatic train operation (ATO) to reduce power consumption and improve ride comfort.
• Predictive infrastructure maintenance control (Predictive Maintenance): AI analyzes vibration/temperature/current data from IoT sensors attached to key railway infrastructure such as switches, signals, and overhead catenary lines to predict in advance when failures will occur. The control center can dispatch maintenance teams before a fault occurs and prevent service interruptions.

Vision AI and sensor-based passenger behavior analysis

• Instant detection of abnormal behavior and dangerous situations on platforms: Applies action recognition (Action Recognition) vision AI to CCTV video installed in stations. The system immediately recognizes passengers falling onto the tracks, prolonged loitering, staggering drunks, violence, and medical emergencies (collapses), and sends pop-up alarms to the control center and station staff smartphones.
• In-station crowd density forecasting and flow control (Crowd Flow Simulation): Fuses 3D LiDAR cameras, smartphone Wi-Fi signals, and ticket gate passage data to visualize real-time congestion inside the station as a 3D heat map. When signs appear that crowds are concentrating in a particular transfer passage or staircase, it automatically guides people along detour routes via digital signage (information displays) to prevent crowd crush accidents.
• Pattern analysis for abnormal ticket use and fare evasion detection: Analyzes transit card tap data patterns as time series to find patterns that deviate from normal commuting routes, such as abnormally passing through gates repeatedly in a short time or improperly using discount tickets, and sets sections for intensive ticket inspection.
• Identification of mobility-impaired passengers and linkage to tailored services: AI recognizes wheelchairs, strollers, white canes for the visually impaired, and the like, to prepare wheelchair lift operation or instruct station staff to provide guidance in advance, shortening travel time for mobility-impaired passengers.

Digital twin integrating control and passenger data (Digital Twin)

• Integrated delay ripple effect simulator: Simulates on the digital twin what chain of ripple effects (Ripple Effect) a 10-minute train delay at Terminal 1 has on passenger density and waiting lines at a transfer station an hour later. Going beyond simply sending trains out quickly, the system issues control commands while also judging "Can that station's platform hold this many people?"

Major railway data providers and items

• Ministry of Land, Infrastructure and Transport (TAGO) Train Information API: Provides in real time train timetables by departure/arrival point including KTX, vehicle types, and lists of trains and city codes by province. - https://www.data.go.kr/data/15098552/openapi.do
• KRIC Rail Portal: Provides train operation status and station operation information for urban and metropolitan railways nationwide (Seoul Capital Area, Busan, Daegu, Gwangju, Daejeon). - https://data.kric.go.kr/rips/index.jsp
• Korea National Railway station information: Provides data on location (latitude/longitude), station class, related lines, number of train stops, and services handled for railway stations nationwide.
• Korea Railroad Corporation (Korail) public data: Provides freight transport fare information, boardings and alightings by station, platform screen door and vertical transport facility status, freight train timetables, and more.
• Seoul Metro train operation status: Provides data on Line 1–8 sections, operating distance, number of stations, travel time, and number of train sets owned.

Key platforms available

• Public Data Portal: Search for various structured data files and APIs such as railway operation information and station information
• KRIC Rail Portal: Use real-time data and statistics by railway operator
• National Transport Data Open Market: Platform for distributing, buying, and selling transport-related data

Ranking of overseas railway AI dispatch-assist solutions

Ranking of overseas railway AI dispatch-assist solutions - Survey of reference models for the "AI train traffic control and passenger behavior analysis" item for supply to KORAIL (survey date: 2026-10-21)

Overview

This page lists overseas cases of AI-based dispatch-assist solutions (delay ripple prediction, recovery scenario recommendation, passenger flow integration), found by search and ordered by usefulness as reference models for the item to be supplied to Korail. Most figures are self-reported by operators or vendors, and no independent verification was found.

Ranking criteria

• Is AI the core - Whether machine learning or reinforcement learning, not rule-based methods or operations research (OR), is the core of the product
• In live operation - Whether it goes beyond research and pilots and is actually used in control centers
• Is it advisory - Whether it is decision support (Decision Support) that leaves the final decision to the dispatcher
• Passenger layer integration - Whether passenger flow and congestion forecasts are integrated into dispatch decisions
• Reproducibility - Whether the structure is one a small or medium-sized supplier could follow through co-development with an operator

Ranking summary table

1. Deutsche Bahn KI-Disposition

• Country / Operator: Germany, Deutsche Bahn. Started at S-Bahn Stuttgart, expanded to München and Rhein-Main; trial use outside the closed S-Bahn system on the Elmshorn–Sylt line
• Description: An AI assist system that gives action recommendations to dispatchers (Disponent). Simulates multiple control variants at about 100 times real-time speed, which can be scrubbed back and forth like a video
• Results (DB announcement): During disruptions, punctuality improved by up to 3 percentage points and up to 8 minutes of delay absorbed. Claims double-digit improvements possible on large networks
• Implications: Developed in-house by the operator, so no vendor lock-in. The prototype of the "AI advises, humans decide" structure
• Sources: DB official, CIO.de, it-daily

2. JR West × alt railway dispatch assist AI

• Country / Operator: Japan, JR West (JR West Innovations invested in alt in 2020)
• Description: Co-development that digitizes veteran dispatchers' traffic rescheduling judgments. Entry into the product development stage announced in May 2023
• Results (announced): Across dozens of past cases, the AI's plans were equal to or better than the dispatchers' plans. The goal is to reduce variation in rescheduling quality due to differences in dispatcher experience
• Status: Commercialization in 2025–2026 unconfirmed. Separately, in May 2026 JR West began a field trial with Weathernews of generative AI support for service suspension decisions, aiming for full operation in fiscal 2027
• Implications: The startup–operator co-development structure is the closest to Korail's benefit-sharing scheme model
• Sources: THE BRIDGE, Kyodo News PR Wire, Kumamoto Nichinichi Shimbun

3. Hitachi AI traffic rescheduling (JR Kyushu field trial) and Hitachi Rail new TMS · HMAX

• Country / Operator: Japan, JR Kyushu (project launched January 2021, verification on actual lines from April)
• Description: Combines machine learning with railway operation management AI to automate traffic rescheduling when the timetable is disrupted. Verified rescheduling on par with skilled dispatchers and faster drafting of timetable change plans
• Follow-up: At InnoTrans 2026, announced a new TMS that uses AI to detect and resolve complex conflicts and an ATS that adjusts operations based on demand. In 2025, launched HMAX, an edge AI platform based on NVIDIA IGX, co-created with Tobu Railway
• Results: No figures published since 2021
• Implications: The global competitor most likely to enter the Korean market
• Sources: Hitachi 2021 press release, Hitachi Rail InnoTrans 2026, Hitachi Rail HMAX

4. MTR Hong Kong passenger demand forecasting model + Intelligent Crowd Diversion System

• Country / Operator: Hong Kong, MTR (jointly with HKUST)
• Description: Simulates crowd scenarios using public transportation and MTR passenger pattern data to decide how many extra trains to add and how to deploy station staff. Claims about 90% accuracy. Operated ICDS, which predicts platform waiting times using CCTV AI people counting, at more than 20 events at Kai Tak Stadium
• Status: In live operation as of August 2025, with plans to strengthen real-time capability by shortening processing time
• Implications: The clearest live case of "passenger behavior analysis → dispatch decisions"
• Sources: SCMP, RTHK, HKIE Transactions

5. Alstom Mastria

• Country / Operator: Panama Metro (2020)
• Description: A multimodal supervision and orchestration product that fuses train load sensor, ticketing, signaling, CCTV, and mobile network data to predict passenger distribution and recommend added services, passenger dispersal, station entry restrictions, and dispersal on platforms. Four functions: multimodal supervision, traffic management, operations coordination, and predictive analytics
• Results (Alstom announcement): Predicts local saturation 30 minutes ahead, 12% reduction in waiting time
• Status: No further adoption confirmed since 2020
• Implications: The only commercial case integrating dispatch and passenger analysis in a single product
• Sources: Alstom 2020, Railway Age

6. Wabtec Movement Planner (formerly GE Transportation)

• Country / Operator: US freight railroads, Norfolk Southern, etc. (installed at 2 companies, in progress at 2)
• Description: An operations planning system based on three-stage optimization: network visualization, network optimization, and automatic routing. Took more than 10 years to develop
• Results: Norfolk Southern average speed up 2 mph (INFORMS 2017). Vendor claims 10% increase in network velocity and 50% reduction in crew hours-of-service expirations. Product revenue over USD 100 million
• Implications: The only case where dispatch optimization has been proven in revenue. However, it is closer to operations research (OR) than AI and focuses on freight
• Sources: INFORMS Interfaces, Wabtec, INFORMS case

7. Resonate Luminate

• Country / Operator: UK, Network Rail. Deployed on the Western route in 2018; expanded in 2022 to Anglia, Western, the Elizabeth line, and Wales under a 7-year contract
• Description: A TMS that predicts the onward paths of each train type to flag conflicts and delays in advance and supports real-time replanning. Ran a £4 million pilot project integrated with GWR's crew and rolling stock systems
• Contract model: In the first pilot (Paddington–Bristol), Resonate bore most of the cost and shared the delay savings (the reduction in compensation payments to operators) with Network Rail. Target of 15% reduction in reactionary delay (not a measured figure)
• Implications: The performance-linked contract structure fits Korail's preference for its benefit-sharing scheme
• Sources: Rail Technology Magazine, Find a Tender 2022, Rail Forum

8. SBB Rail Control System (RCS-Dispo · RCS-ADL)

• Country / Operator: Switzerland, SBB (developed and operated in-house since 2009), BLS · SOB (in use since 2014)
• Description: Every 2 seconds, predicts the state of the entire Swiss network up to 90 minutes ahead, automatically detects occupation conflicts, and proposes solutions. The ADL module sends energy-optimal driving recommendations to more than 1,500 trains a day (saving 136,000 kWh)
• Is it AI: Algorithmic prediction and rule-based proposals, not machine learning. Host of Flatland, a reinforcement learning rescheduling challenge
• Implications: The textbook "prediction–conflict–advice" architecture
• Sources: Global Railway Review Q&A, Global Railway Review

9. Thales NetTrac ARAMIS (now Hitachi GTS)

• Country / Operator: 18 customers in Germany, Israel, Austria, Finland, Denmark, etc. (vendor claim)
• Description: Capacity optimization through automatic conflict detection and operational solution proposals. Rule-based TMS
• Implications: The off-the-shelf product to be faced in Korail control center replacement projects
• Sources: Thales

10. Beijing Railway Bureau RTARS

• Country / Operator: China, Beijing Railway Bureau
• Description: Real-time automatic timetable rescheduling in a human–computer interaction framework where the dispatcher proposes adjustments and the system recalculates. A 2023 High-speed Railway paper reports an actual implementation
• Status: No official announcement of nationwide deployment confirmed
• Sources: TechXplore

11. Trafikverket MATRIX

• Country / Operator: Sweden, Trafikverket · Blekinge Tekniska Högskola
• Description: Uses decision trees and random forests to suggest which train gets priority in a conflict, with the dispatcher making the final decision. Final report in January 2024
• Status: Experimental research, not deployed
• Sources: MATRIX final report, BTH

12. Research-stage group

• NS · ProRail ICAI RAIL Lab / Utrecht AI & Mobility Lab (Netherlands): Prototypes of planning support tools that people can trust, research on human–AI collaborative decision-making. Position that full automation is undesirable - ICAI RAIL lab, Utrecht AI & Mobility Lab
• SNCF REINFORCERAIL (ANR, France–Germany) · SNCF–CEA collaboration: Uses neural networks to select which trains to adjust and the adjustment time window during disruptions. Preliminary results show feasibility; model complexity and training burden noted as issues - ANR REINFORCERAIL
• Flatland · DISPLIB: A reinforcement learning rescheduling benchmark and a library of train dispatching problems - DISPLIB

Implications for the Korail item

• All successful cases are advisory: DB, alt, MATRIX, and RTARS all left the final decision to the dispatcher. Hitachi and JR Kyushu, which promoted fully automatic traffic rescheduling, have not released performance figures since 2021.
• Metro operators lead on the passenger layer: MTR and Mastria are the only live cases, and no case was found of a mainline railway integrating passenger flow into dispatch decisions. Demonstrating this integration first on Korail's metropolitan commuter rail would be a differentiator.
• Use the Resonate approach as a reference for the contract model: A structure where the supplier bears the initial cost and shares the delay savings fits Korail's preference for its benefit-sharing scheme.
• Competitive landscape: Hitachi (HMAX · new TMS) and Thales/Hitachi GTS (ARAMIS) are the incumbent competitors in Korail control center projects.

Research limitations

• Siemens Mobility mentioned "AI-based disruption management" at InnoTrans 2026, but no dispatch advisory product was confirmed.
• The figures for DB, alt, MTR, Alstom, and Wabtec are all self-reported by operators or vendors, and no independent verification was found.
• Whether the alt · JR West product was commercialized in 2025–2026, the DB system's development partners, the customers of Hitachi's new TMS, and whether China has deployed nationwide remain unconfirmed.
• Huawei's 2025 intelligent dispatching announcement page and the original SCMP article could not be checked directly due to access restrictions (HTTP 403).

See also

• KORAIL
• Blackhole