Propagating agents with macroscopic dynamic network loading Challenges and possible solutions ir. Jeroen van der Gun dr.ir. Adam Pel prof.dr.ir. Bart van Arem
Dynamic network loading (1) Simulating traffic propagation over time Uses graph of links and nodes Needs to know amount and routing of traffic Time
Dynamic network loading (2) Microscopic (agent-based) Vehicles are agents Traffic flow emerges from their interaction, including complex phenomena like hysteresis Macroscopic (aggregate) Parameters typically more easily observable Shorter computation time (lower complexity) Getting better at reproducing complex traffic phenomena
Agent-based demand Activity-based modelling Sampled choice behavior Agents can adapt their initial plans to disruptions and emergencies Sampled choice behavior Agents can retain consistent attitudes over time Multimodal transportation systems Passenger agents can board and alight public transport vehicles, which are also agents
? Four combinations Aggregate demand Agent-based demand Macroscopic network loading Aggregate demand is fed directly into the macroscopic network loading model ? Microscopic network loading Aggregate demand can be split into discrete vehicle agents Decision-making agents map one-to-one to vehicle agents Aggregating agent-based demand precludes agent-level decision-making while en-route
Contours of a solution (1) Macroscopic dynamic network loading models have multi-commodity formulations Can be used to distinguish individual agents in the macroscopic flow of traffic
Contours of a solution (2) Individual agents also need an unambiguous location Agent location is defined as the “front” of its vehicle “Tail” of its vehicle must follow this “front” I always have an intended next turn too
Pick a suitable network loading model
Link modelling challenges Avoid systematic errors Adherence to LWR varies across models Limit numerical diffusion Variational methods perform best due to fewer discretizations Prevent agents from traversing links faster than the free speed Problem for models with discrete time unless the time within a time step is explicitly considered Accept agents to flow into a link Computational burden or accuracy problems for models with discrete vehicle count if the number of agents is large
Turning fraction challenges Specify the ordering of vehicles on a link Need a strict weak order of vehicle parts to avoid highly fluctuating discrete turning fractions Otherwise, e.g. an entire motorway is blocked while one vehicle is moving into an off-ramp For models with discrete vehicle counts this implies individual lanes need to be considered Choose a node model of an appropriate form Incremental formulations have less vehicle diffusion than squeezing formulations Prevent “agent-based gridlock” The “tails” of agents must not be able to block each other
preserving agent-based en-route choices Conclusions Aggregate demand Agent-based demand Macroscopic network loading You can go here preserving agent-based en-route choices but be careful Microscopic network loading Discrete vehicle count may not make things easier Link Transmission Model appears reasonably able to deal with the challenges