In any field the use of terms to define a certain approach is a matter of dispute. There are mulitple classifications which are sometimes overalapping with each other. For instance, in traffic flow, I have seen researchers classifying into either macroscopic or microscopic. Some of them add the mesoscopic models between them. Others add the network fundamental diagram (NFD) approach as a distinct one, but its relationship are similar to macroscopic models… Anyway, it is not the point here to add the classification I prefer into the topic.
Rather, I will talk about agent-based models (ABM) in transportation. The agent-based model approach first came from computer science to describe models in which complex systems were modeled by individual agents that have their own attributes takes decision based on their perceived state of the environment and their decisions also impact the enviroment itself. In Transportation Systems there is an obvious agent type to be modeled: a person. This person may engage activities and undertake trips to perform in these activities. However, in transportation we can think in several other agents that takes decisions such as vehicles, traffic signal controllers, ramp meters, to name a few.
In summary, I would describe agent-based models in transportations models in which the system is modeled as multiple agents with individual behavior, with a traveler or person being one among multiple agent tyypes, and the outcomes are based on the emergent outcome that results from the agents interaction. The key distinction with their counterparts such as aggregated and macroscopic models is higher level of details on each metric and especially variation within the population. For example, with four step models, one of the outputs is the number of trips (demand) and the travel time for a given origin destination pair. In agent-based models, the number of trips is based on their agent choices and we can, for example, determine the number of trips per each activity type (work, personal activity, leisure, etc.) and the distribution of travel time for each activity type (perhaps work trips are concentrated in some periods so might be different from the others). This is an example, but I think it is possible to have an idea. We could also extract mode information for the work trips too.
Commonly, these tools features a traffic simulation and a travel behavior with the agents that take decision such as:
- Mode
- Destination
- Route
- Departure Time
- Vehicle purchase
- Workplace location
There are tools that follows this approach, here are some of them:
- POLARIS (I am a core developer so may be biased on it :-))
- MatSIM
- SimMobility
As a conclusion, I will leave a personal disagreement with the term. In my opinion, it is hard to distinguish between microscopic and agent-based models. As I understand, the distinction would be in the “intelligence” of the agents that may evolve and take “smarter” decisions over time. From this standpoint, I believe that most of the models within transportation I would name as microscopic models. Since it is not me that determine names and that agent-based sells more, probably I will keep using the agent-based term for a while.
If there are agent-based model tools that you want to be added to this list, please contact me.