Dear colleagues,
System Dynamics is known as a rationale to examine systems presenting characteristics of dynamic complexity.
It is composed of two modelling paradigms in System Dynamics:
- Causal Loop Diagrams (CLDs): useful in representing the fundamental feedback structures of a system and qualitatively assess its potential behaviours; and
- Stock and Flow Diagrams (SFDs): that enables quantifying the relationships among variables and determine their accumulation and depletion over time through simulation.
There seems to be a consensus in the area that the use of these complementary paradigms should help to better understand the reasons for a given problematic behaviour.
Still, the way to use the two modelling paradigms seems to differ. In some cases, researchers make use of the CLDs to explore a problem, structure the multiple factors influencing it and speculate behaviour from the structure. In other cases, researchers directly dive into defining a scope of a system to investigate, determine SFDs structures and experiment with the system against reference modes to critical variables.
I have heard the terms problem-oriented and system-oriented modelling. Or perhaps the remarks: "By doing that, you are modelling the system and not the problem".
The following questions emerge: How to relate these terms (problem-oriented vs. system-oriented) to the available modelling options? Is there a rationale to choose an appropriate approach to a given situation? Is there any discussion in the SD literature or other areas that might shed light on this issue?
Many thanks in advance,
Daniel!
Daniel Guzzo I'm not an expert in this specific area, but a useful insight I learned with respect to modelling/simulation is that an in silico model or simulation should not match data, because at that point you would have fitted to data, with no generalization capability (and unlike to distillate the causal processes underlying the data).
A model (problem) based approach (in my limited experience), allows you to quickly verify where the description of your model diverges from data, e.g. you test a hypothesis and are able to some extent to invalidate (closed form) assumptions.
Simulation (system) (can) work(s) in the other direction, where the outcome of the system at time t allows you to discover unknown aspects, in addition to going beyond feasibly specified models.
They meet in the middle where you compute a surrogate function from a simulation that specifies a model (roughly).
Not sure if this is of help, or correct,
Ben
Example (system analyses and dynamic control): PREFERENCES ANALYTICAL REPRESENTATION AND VALUE DRIVEN CONTROL AND MANAGEMENT OF COMPLEX SYSTEMS: UTILITY APPROACH in Rresearchgate
Ben Cardoen
That is a rather insightful response.
Putting too much effort into making a simulation model (only) fit data will not help to gather insights.
On the other hand, calibration to existing data is important to enhance model validation and trust in those 'unknown aspects' discovered.
I am sure it helps in the discussion. And it is cool to gather the perspective from a different modelling approach.
Daniel!
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling