This is machine translation

Translated by Microsoft
Mouseover text to see original. Click the button below to return to the English verison of the page.

Note: This page has been translated by MathWorks. Please click here
To view all translated materals including this page, select Japan from the country navigator on the bottom of this page.

Optimization Does Not Make Progress

Should I worry about the scale of my responses and how constraints and design requirements are discretized?

No, Simulink® Design Optimization™ software automatically normalizes constraints, design requirement and response data.

Why don't the responses and parameter values change at all?

The optimization problem you formulated might be nonsmooth. This means that small parameter changes have no effect on the amount by which response signals satisfy or violate the constraints and only large changes will make a difference. Try switching to a search-based method such as simplex search or pattern search. Alternatively, look for initial guesses outside of the dead zone where parameter changes have no effect. If you are optimizing the response of a Simulink model, you could also try removing nonlinear blocks such as Quantizer or Dead Zone.

Why does the optimization stall?

When optimizing a Simulink model, certain parameter combinations can make the simulation stall for models with strong nonlinearities or frequent mode switching. In these cases, the ODE solvers take smaller and smaller step sizes. Stalling can also occur when the model's ODEs become too stiff for some parameter combinations. A symptom of this behavior is when the Simulink model status is Running and clicking the Stop button fails to interrupt the optimization. When this happens, you can try one of the following solutions:

  • Switch to a different ODE solver, especially one of the stiff solvers.

  • Specify a minimum step size.

  • Disable zero crossing detection if chattering is occurring.

  • Tighten the lower and upper bounds on parameters that cause simulation difficulties. In particular, eliminate regions of the parameter space where some model assumptions are invalid and the model behavior can become erratic.

Was this topic helpful?