For years, combined transport operators have been asked to do the impossible: move more freight from road to rail without a corresponding expansion of infrastructure. Train paths are scarce, construction sites are frequent, and customers still expect fast, reliable door-to-door services.
In this light, the German research project KIBA – short for Artificial Intelligence and Discrete Loading Optimisation Models for Enhanced Utilisation in Combined Transport – set out to answer a very practical question: how far can artificial intelligence and mathematical optimisation push today’s rail networks within the existing physical boundaries?
One of the core answers KIBA delivers is a new approach to network planning optimisation for combined transport.
The planning pain behind today’s intermodal networks
In many intermodal networks, planning still reflects the constraints of the past. Capacity is often planned per relation, per terminal or per individual train. Planners work with spreadsheets, static capacity tables and a lot of experience. That experience is valuable – but it has limits once you try to coordinate hundreds of relations, dozens of hubs and a constantly changing demand pattern.
At the same time, demand is volatile, and disruptions are part of daily life. Customers book late, cancel or upgrade loads at short notice; infrastructural works and operational delays force last-minute changes. Trying to keep a European-wide network balanced under these conditions, with only manual planning, inevitably leaves potential capacity on the table and makes it harder to keep rail competitive against road.
KIBA in a nutshell
KIBA is a multi-year research and innovation project funded by the German Federal Ministry for Digital Transformation and Government Modernisation. Led by Kombiverkehr KG as combined transport operator and network coordinator, the consortium brings together Deutsche Umschlaggesellschaft Schiene-Straße (DUSS) as terminal operator, VTG as wagon and railcar specialist, INFORM as optimisation and AI software provider, Goethe University Frankfurt and the Technical University of Darmstadt as academic research partners, and KombiConsult as consulting firm for combined transport.
The goal: develop and test a demonstrator for a capacity management and loading optimisation system in combined transport. Conceptually, it consists of three optimisation modules:
1. Wagon-level loading
2. Train-level loading
3. And network-level utilisation
This article focuses on the third component: network planning optimisation – deciding which load units should travel on which trains, along which routes and at which times, to make the best possible use of the available network. Further below, we also provide an insight into how network planning optimisation interfaces with train load optimisation.
What network planning optimisation actually does
From an operational perspective, network planning optimisation in KIBA addresses a simple but powerful question:
Given the capacities of trains for their particular routes over the next days and weeks – how should all current and expected load units be routed through the network to achieve the best overall result?
Instead of planning train by train or terminal by terminal, network planning optimisation looks at the entire system. It decides:
— How much capacity on each train is used for which origin–destination relations
— Which paths through hubs and corridors should each load unit (or class of units) take
— How to respect time promises and capacity limits while keeping flows balanced
The objectives reflect what operators and their customers care about: maximise utilisation and transported volume, minimise transport time and cost, and avoid overloads, congestion and unnecessary re-handlings at hubs.
The data foundation: forecasts meet bookings
KIBA builds on three pillars for the network planning optimisation:
1. A shared view of the network
A central description of wagons, trains and infrastructure forms the backbone of the system: maximum weight and length, loading length and slots per equipment type, valid routes, hubs and capacity constraints.
2. Demand forecasts on the network
Machine-learning models analyse historical booking data and derive forecasts for future load-unit volumes per origin–destination pair, weekday, equipment type and weight category. These forecasts are exported in standardised formats so that they can be ingested directly by the optimisation software and combined with real bookings.
3. Current bookings and schedules
Operational data from the combined transport operator – orders, timetables, connections and wagon patterns – are supplied via established interfaces such as EDIGES.
In the KIBA demonstrator, a central controller gathers these inputs, combines them with static network and timetable data, and then triggers the appropriate optimisation runs.
How the optimisation runs in practice
For a combined transport operator, network planning optimisation in KIBA supports two main modes of use.
Mode 1: Tactical weekly planning
First, there is a tactical planning run over a horizon of several days or a week. The system takes:
- Confirmed bookings for the planning period
- Demand forecasts for the considered horizon
- And the current timetable and available trainsets
Based on this, it computes how many units of which type should be routed via which trains and hubs. It respects:
- Maximum train length and weight
- Available slots per segment and per unit type
- Latest delivery times at the destination
- And logical routing constraints, such as mandatory hubs or forbidden paths
Runtimes support regular planning cycles, rather than just one-off studies every few months.
Mode 2: Short-notice re-optimisation
Second, freight does not always behave as planned. New bookings arrive, trains are cancelled, and infrastructural works overrun. For these situations, KIBA provides fast re-optimisation runs:
- New bookings without an assigned train can be evaluated against the current plan to find feasible and convenient routings.
- If a route is blocked or a train is cancelled, affected units can be rerouted to alternative trains and paths while still respecting customer deadlines and capacity limits.
From an operations-centre point of view, this turns the static timetable into a living network plan that can be updated as conditions change.
Interfacing with train loading optimisation
Network planning optimisation in KIBA does not exist in isolation: its decisions can be refined even further at a train and wagon level.
Upstream, the network optimiser works with aggregated capacity – length, weight and slots on each train and segment. Downstream, the train-loading optimiser takes the result for a given train – the list of load units assigned to it – and computes a detailed plan that places each unit on a specific wagon position while satisfying all technical loading rules and safety regulations.
By feeding relevant wagon and loading constraints back into the network level, KIBA ensures that the “perfect” network plan is not one that later turns out to be impossible to load in practice.
What kind of gains are realistic?
Tests with real data show that network-wide optimisation can significantly increase average utilisation of available train paths and wagons, because it aligns expected and actual demand with capacity across the full network rather than train by train.
At the same time, the optimisation model actively penalises unnecessary re-handlings and late deliveries at the destination. That steers flows away from overloaded hubs and fragile routings, helping to reduce congestion and delays.
In disruption scenarios, the ability to recompute routings quickly supports a more controlled reaction: instead of ad-hoc problem-solving, planners see consistent proposals that preserve service levels as far as possible and make transparent where compromises are unavoidable.
In validation with historical data, the optimised plans were nearly identical to real-world decisions made by experienced planners and revealed additional capacity or more robust routings that would be hard to identify manually in a complex network.
For an industry trying to shift more cargo to rail without waiting for new infrastructure, these are exactly the types of gains that matter: more volume, more stability, more transparency – with the assets already available.
Looking ahead: corridor-level potential
So far, KIBA focuses on a single operator’s network. But the logic extends naturally to international corridors and multi-operator contexts. If partners are willing to share at least aggregated capacity and demand information, similar optimisation approaches could support:
- Coordinated capacity planning on key corridors
- Proactive handling of major disruptions
- And better alignment between infrastructure works and freight flows
For an industry under pressure to deliver more with less, network planning optimisation of the kind explored in KIBA is not just a research curiosity. It is a concrete step toward turning European combined transport networks into living, data-driven systems that respond intelligently to demand and disruption – and make rail a more attractive default choice in logistics.