A model-based adaptive planning framework using surrogate modelling for urban transport systems under uncertainty

Adaptive planning helps cities adapt to an uncertain future environment by providing guidance on the required interventions, conditional on how the future evolves, to best achieve planning goals. Such plans can be identified through detailed agent-based models, but they are usually computationally expensive, limiting their ability to run repeatedly for multiple scenarios. This paper proposes a framework for developing adaptive plans for urban transport systems using surrogate models (i.e., fast approximations) of detailed models to determine which adaptive plans are best achieving multiple objectives for a large number of possible future scenarios. For the first time, the framework enables urban-scale simulations that are sufficiently fast that they can be used in real world planning processes, to evaluate large ensembles of different combinations of future scenarios and urban interventions. The framework is empirically validated with an agent-based urban transport model (MATSim) of a Singaporean neighbourhood to investigate interventions to be implemented in response to autonomous vehicles (AVs) development. A surrogate model reduces computational time by five orders of magnitude, at the expense of reduced accuracy. Adaptive plans are then developed as sequences of interventions (e.g., network reconfiguration), triggered over timebased on indicators (e.g., travel demand) to achieve multiple planning goals (e.g., reduction in carbon emissions). The framework allows the interaction between planners and stakeholders to evaluate the whole set of potential plans and build consensus towards a best plan.