Customizable Contraction Hierarchies for Mixed Fleet Vehicle Routing : Fast weight customization when not adhering to triangle inequality

Abstract: As the transport industry shifts towards Battery Electric Vehicles (BEVs) the need for accurate route planning rises. BEVs have reduced range compared to traditional fuel based vehicles, and the range can vary greatly depending on ambient conditions and vehicle load. Existing research focuses more on the theoretical algorithms, and often have none or very simple vehicle models, leaning towards consumer cars instead of heavy duty trucks. Vehicle Route Planning (VRP) is a wide research area, and this thesis focuses on the Shortest Path subproblem. Contraction Hierarchies (CHs) is a commonly used family of algorithms for finding shortest paths in road networks, and is prevalent in the research frontier. CHs however comes with certain drawbacks, such as having to perform a costly preprocessing phase whenever metrics change, and not being able to share map data between multiple vehicles in a fleet. This thesis extends CHs to support a mixed fleet, with fast metric updates and support for more detailed cost optimization goals. This is done by implementing Customizable Contraction Hierarchies (CCHs), but with custom data structures and customization phase. This implementation allows map data to be shared between vehicles in a fleet, and keeps each vehicle's edge weights separate. The edge weights can be updated quickly, as the customization phase scales linearly with the size of the map. The implementation also supports edge weights that do not adhere to triangle inequality, which the previous research did not. Experiments are executed on a map of Stockholm and a synthetic map, to test the algorithm's performance, verify correctness, and stress the importance of accurate metrics for optimization goals. The CCH performed as expected, if not better, and its correctness is upheld. The implementation is fit to be integrated into a route planner, but further research should be conducted to see how it meshes with other parts of VRP, such as time windows, turn costs, and charging stations.