Mobility-as-a-service personalised multi-modal multi-objective journey planning with machine-learning-guided shortest-path algorithms

Mobility-as-a-service (MaaS) apps provide a single platform for journey planning, booking, payment and ticketing, and are proposed as a medium for encouraging sustainable travel behaviour. Generating sustainable-vehicle-based journey alternatives can be formulated as a multi-modal multi-objective journey-planning problem, one that is known to have a prohibitively large solution space. Building on prior insights, we develop a scalable decomposition-based solution strategy. A Pareto set of journey profiles is generated based on inter-transfer-zone objective criteria contributions. Then, guided by neural-network predictions, extended versions of existing shortest-path algorithms for open and public transport networks are used to optimise the paths and transfers of journey profiles. A novel hybrid k-means and Dijkstra’s algorithm is introduced for generating transfer-zone samples while accounting for transport network connectivity. The resulting modularised algorithm knits together and extends the most effective existing shortest-path algorithms using neural networks as a look-ahead mechanism. In experiments based on a large-scale transport network, query response times are shown to be suitable for real-time applications and are found to be independent of transfer-zone sample size, despite smaller transfer-zone samples, leading to higher quality and more diverse Pareto sets of journeys: a win-win scenario.