A closed loop between prediction and decision.
TRANSFORM uses a single estimate-then-optimise framework that couples predictive models with optimisation under uncertainty, closing the loop so that operations and planning inform one another.
Estimate, then optimise, in a loop.
Prediction of disruption and uncertainty and optimisation for planning have developed as separate research streams. TRANSFORM combines them.
A probabilistic forecasting method integrates directly into optimisation without violating linearity, while quantifying uncertainty in a dynamic environment. Predictions inform optimisation, realised outcomes correct the predictions, and the loop adapts as new information becomes available.
From open loop to closed loop.
Prediction, behaviour and optimisation are often treated as separate problems. Three limitations recur:
Behavioural models without adaptation over time; choice parameters stay fixed as the network changes.
Predictions that do not inform optimisation; forecasts produced separately from the decisions they should guide.
Optimisation without foresight, at a single timescale; plans that do not adapt to daily conditions.
Unravelling the unknown, and making educated decisions for the long term.
TRANSFORM combines the three into one adaptive, multi-timescale approach. Predictions are made uncertainty-aware and optimisation-ready, behaviour is allowed to adapt, and the result of each decision becomes an input that informs the next.
The four themes.
Quantification & probabilistic forecasting
A two-stage approach provides short-term probabilistic predictions of resource availability. The first stage estimates hourly availability from weather and demand; the second calibrates against recent observations. The forecasts are formulated so they can be embedded in optimisation.
Behavioural estimation & adaptation
Choice parameters are estimated for the current system and updated when new mobility solutions are introduced, without re-running surveys. Overlapping services are handled by enriching the data; non-overlapping services through simulation-based population synthesis.
Optimisation across time horizons
Assortment optimisation is used to design a multimodal trip menu: a subset of options that reflects each traveller's preferences and real-time fleet availability. Linearisation keeps the problem tractable in real time and across timescales.
Multi-agent simulation: TRANSSim
TRANSSim extends the existing MaaSSim platform with network adaptation and a coordinated multimodal trip recommender, and adds temporal aspects so that agents at different time scales remain coherent. A discrete-event scheduler processes events in chronological order.
Work packages
How the research is organised.
Three thematic packages cover supply, demand and network adaptation. A cross-cutting package provides the shared uncertainty quantification and the simulation testbed.
Uncertainty quantification & impact assessment
WP0 supports every other package and runs across the whole project. It identifies the sources of uncertainty (systemic, environmental and behavioural), drawing on multi-source data from the Netherlands, Germany and Norway, and builds TRANSSim by extending the MaaSSim platform.
Characterise and quantify multi-source uncertainty and how it grows across timescales.
Extend MaaSSim to a multi-timescale, multi-agent platform for impact evaluation.
Risk & mitigation. Agents at different time scales are hard to keep coherent; a discrete-event scheduler processes events in chronological order regardless of native scale.
Multimodal supply management under uncertainty
WP1 addresses short-term supply: the fleet of modalities across the network. It proposes a probabilistic resource-availability prediction that integrates cleanly with optimisation, then a real-time framework for centralised asset management. Its outputs feed WP2 at the same temporal scale.
A two-stage Bayesian-inspired model: hourly availability from weather and demand, calibrated against recent observations.
Integrate probabilistic predictions with service-scheduling optimisation that adapts as information unveils.
Risk & mitigation. Where data is unavailable for some modes, simulated and open-access data validate the models.
Demand management & behavioural adaptation
WP2 manages short-term demand through a user-centric multimodal trip recommender. It estimates choice parameters and calibrates them as new mobility solutions appear, without re-running surveys, then designs a menu that respects preferences while accounting for WP1's supply predictions.
Update choice parameters via LLM-based data enrichment, or simulation-based population synthesis for non-overlapping services.
Assortment optimisation designs the trip menu; piece-wise linear functions keep the choice models tractable.
Risk & mitigation. Coupling behaviour with predictions adds nonlinearity; linearisation and smart search keep it solvable in real time.
Long-term network adaptation & expansion
WP3 bridges short-term management to long-term decisions: adaptation, as new services enter an existing network, and expansion, as cities grow. It integrates domain knowledge with Bayesian theory to predict delay propagation across modes, including new services with limited data, feeding interpretable predictions into strategic optimisation.
Interpretable probabilistic predictions of causal effects between modes, combining historical data with limited new-feature information.
An adaptive framework that embeds uncertainty quantification of new services into strategic investment decisions.
Risk & mitigation. Validation uses an already-expanded city with before-and-after scenarios; data access is secured with the Dutch Ministry of Infrastructure and Water Management.