In modeling and simulation theory, an ontology is a formal description of the objects and relationships that a computer system intends to simulate. The term “ontology” comes from technical philosophy, where it refers to the structure of what exists. Though it may superficially appear that an ontology is just a description of a model’s data structures, there is a crucial difference: data structures are manipulated inside the model, whereas the ontology makes a stronger claim about what exists “outside” the model. Effective ontologies are essential for a variety of practical modeling activities that involve the communication of data and information, such as deriving model inputs from “big data” streams, passing information from one model to another, developing meaningful performance measures, and communicating model results to stakeholders and the public. In transportation modeling, practitioners rely heavily on common sense about the “real world” to supply an ontology. But that practice has a number of pitfalls that can lead to inefficiencies in model design, inconsistent outputs, and errors in calculating and interpreting performance measures.
Key challenges in establishing an effective ontology will be addressed by exploring and systematizing the ontologies that are implicit in trip and activity-based models. The important differences between these approaches to travel modeling can be clearly framed in terms of the objects the models manipulate. Using that “ontological perspective”, it is possible to clarify and effectively address practical issues related to the computation, presentation and interpretation of performance measures. Of course, ontologies do not just apply to models. In the realm of “big data”, having a clear understanding of the objects of study is critical to extracting information that supports effective modeling and analysis. The importance of consistent ontologies in data and models will be illustrated by exploring the ontology of travel time reliability in relation to the data we collect, and how we approach that data when we develop performance measures and models.