Any Logic - Multi-method Simulation Modeling

Multi-method Simulation Modeling

AnyLogic models can be based on any of the main simulation modeling paradigms: discrete event or process-centric (DE), systems dynamics (SD), and agent-based (AB).

System dynamics and discrete event are traditional simulation approaches, agent based is new. Technically, the system dynamics approach deals mostly with continuous processes whereas "discrete event" (by which we mean all descendants of GPSS also known as process-centric simulation approach) and agent based models work mostly in discrete time, i.e. jump from one event to another.

System dynamics and discrete event simulation historically have been taught at universities to very different groups of students, namely management and economy, industrial and operation research engineers. As a result, there are two distinct practitioners' communities that never talk to each other.

Agent based modeling until recently has been mostly a purely academic topic. However, the increasing demand for global business optimization caused leading modelers looking at combined approaches to gain a deeper insight into complex interdependent processes having very different natures.

How modeling approaches correspond to the abstraction levels. System dynamics dealing with aggregates is obviously used at the highest abstraction level. Discrete event modeling is used at low to middle abstraction. As for agent based modeling, this technology is used across all abstraction levels, and agent may model objects of very diverse nature and scale: at the "physical" level agents may be e.g. pedestrians or cars or robots, at the middle level – customers, at the highest level – competing companies.

AnyLogic allows the modeler to combine these simulation approaches within the same model. There is no fixed hierarchy. So, as an example, one could create a model of the package shipping industry where carriers are modeled as agents acting/reacting independently whereas the inner workings of their transport and infrastructure networks could be modeled with discrete event simulation. Similarly, one can model consumers as agents whose aggregate behavior feed a systems dynamics model capturing flows such as revenues or costs which do not need to be tied to individual agents. This mixed language approach is directly applicable to a wide variety of complex modeling problems that may be modeled via any one approach albeit with compromises.