Model structure analysis through graph theory: partition heuristics and feedback structure decomposition

The argument of this article is that it is possible to focus on the structural complexity of system dynamics models to design a partition strategy that maximizes the test points between the model and the real world, and a calibration sequence that permits an incremental development of model confidence. It further argues that graph theory could be used as a basis for making sense of the structural complexity of system dynamics models, and that this structure could be used as a basis for more formal analysis of dynamic complexity. After reviewing the graph representation of system structure, the article presents the rationale and algorithms for model partitions based on data availability and structural characteristics. Special attention is given to the decomposition of cycle partitions that contain all the model's feedback loops, and a unique and granular representation of feedback complexity is derived. The article concludes by identifying future research avenues in this arena.