Tasks

As a first step of a Sensitivity analysis, a model is defined or loaded (using the Model editor). A model is generally defined with a number of input parameters and model outputs where the model outputs are allowed to be time dependent. Different kinds of models are supported by Eikos:

• An expression: a model made up of coupled equations. Each equation is built up of scalar parameters, other equations and mathematical functions (e.g. sin(x), x^y). An expression can also be time dependent by including the reserved variable time.
• An ecolego model.
• An external executable model. Windows .exe and .bat executables are supported. How to access an external model from eikos.

The more general case when output samples can somehow be generated given values of input parameters is also supported (see the section Importing and exporting data).

Each parameter can be assigned one or more of the following (Using the Parameter settings editor):

• A probability distribution function (PDF)
• A scalar value
• A minimum and maximum value

Depending on the selected Sensitiviy Analysis method, either a PDF, scalar value or minimum/maximum value is required. For most methods, probability distributions are required of the parameters. The exceptions are:

• Deterministic: Requires scalar values of the parameters.
• Local Sensitivity Analysis: Requires a scalar, minimum and maximum value of each parameter.
• Simulation Table: Ignores any parameter values and uses values from the simulation table.

For parameters not included in the analysis, scalar values must be set.

Methods supported are

• Probabilistic: Computes correlation measures and the first order sensitivy index (EASI) using random (Monte Carlo or Latin Hyper cube) samples from probability distributions on the model parameters.
• Sobol custom order: Computes a custom order as well as first and total order sensitivity index.
• Sobol group indices: Computes the first and total order sensitivity index of each selected parameter or groups of parameters.
• Extended Fourier Amplitude Sensitivity Test (EFAST). Computes the first and total sensitivity index of each selected parameter or groups of parameters.
• Fourier Amplitude Sensitivity Test (FAST). Computes the first order sensitivity index of each selected parameter.
• Local Sensitivity Analysis
• Random Balance Design (RBD)
• Deterministic: Simply computes each model output at the scalar parameter values.
• Simulation Table: Computes correlation measures and the first order sensitivity index (EASI) using tabulated parameter values.

After defining a model, the model parameters and outputs to be used in the Sensitivity Analysis must be selected in the Parameter selection editor.

Grouping of parameters

Two or more selected parameters can be assigned to a group. In this way, the influence of each of the selected parameters as well as any combination of them is accounted for when computing the sensitivity indices on a given output. The defined groups are only used in the Sobol group indices or EFAST method. For other methods any assigned groups are ignored.