Situation Prediction Nets - Playing the Token Game for Ontology-Driven Situation Awareness

Situation awareness in large-scale control systems such as road traffic management aims to predict critical situations on the basis of spatio-temporal relations between real-world objects. Such relations are described by domain-independent calculi, each of them focusing on a certain aspect, for example topology. The fact that these calculi are described independently of the involved objects, isolated from each other, and irrespective of the distances between relations leads to inaccurate and crude predictions. To improve the overall quality of prediction while keeping the modeling effort feasible, we propose a domain-independent approach based on Colored Petri Nets that complements our ontology-driven situation awareness framework BeAware!. These Situation Prediction Nets can be generated automatically and allow increasing (i) prediction precision by exploiting ontological knowledge in terms of object characteristics and interdependencies between relations and (ii) increasing expressiveness by associating multiple distance descriptions with transitions. The applicability of Situation Prediction Nets is demonstrated using real-world traffic data.