Statistics for Run-Time Control

Some systems rely on statistics to influence the dynamics. For example, a queuing system with discouraged arrivals has a feedback loop that adjusts the arrival rate throughout the simulation based on statistics reported by the queue and server, as illustrated in the Varying Entity Generation Times Via FeedbackVarying Entity Generation Times Via Feedback example.

When you create simulations that use statistical signals to control the dynamics, you must have access to the current values of the statistics at key times throughout the simulation, not just at the end of the simulation. Some questions to consider while designing your model are:

  • Which statistics are meaningful, and how should they influence the dynamics of the system?

  • How can you compute the desired statistics at the right times during the simulation? It is important to understand when SimEvents® blocks update each of their statistical outputs and when other blocks can access the updated values. For more information, see Role of Event-Based Signals in SimEvents.

  • Do you need to account for initial conditions or extreme values in any special way? For example, if your control logic involves the number of entities in a queue, then be sure that the logic is sound even when the queue is empty or full.

  • Will small perturbations result in large changes in the system's behavior? When using statistics to control the model, you might want to monitor those statistics or other statistics to check whether the system is undesirably sensitive to perturbations.

Was this topic helpful?