Vectorial Sensitivity Analysis

This method uses an optimization algorithm to find the values of a set of parameters that maximize/minimize a chosen variable. The method characterizes a given dynamic model as a nonlinear programming problem.

It can be used to analyze how a model reacts to a set of parameters when they perturbed together, instead of analyzing isolated perturbations. This further increases the chances of finding, for example, a combination of small parameter perturbations that has a non-negligible effect on a variable. It can possibly reveal that the model is not robust to some perturbation combinations even when the model is robust against perturbations of the same parameters but tested individually.

Nevertheless, this analysis may take several minutes, maybe hours, depending on the following:

• The complexity of the model. Simpler models take less time to simulate.
• The number of parameters to analyze. Each parameter has to be combined with all the others. This means that no matter how effective the underlying optimization algorithm is, in the worst case scenario the analysis time is affected exponentially by the number of parameters to analyze.
• The simulation time. The shorter the internal simulation time the shorter the execution of the simulations.
• The epsilon value. A smaller epsilon may find a better solution but at the expense of longer optimization analysis time.
• The bounds percentages. The larger the bounds of the parameters, the more values the optimization algorithm must test.
• The system: number of processors, disk speeds, etc.

Also, take into account that if a parameter has no effect on the variable under study then its "optimum" value returned by this feature can be misleading. For each parameter to analyze, it's recommended to check beforehand if it has an effect on the variable. The Individual Sensitivity Analysis, also provided with OMSens, is an useful tool for this verification.

Known limitations

Only parameters and variables of type Real are recognized. Renamings like 'type MyType = Real' are not supported either.

Arrays of any type are also not yet supported.