Parameters
What are parameters in a computational model?
In a computational model a parameter is a variable or input that can be varied to study the behavior of a system being modeled. The system can be anything from a physical process or phenomenon, to a piece of machinery or a biological system. Parameters are critical components of a computational model because they influence the outcome of the simulation.
A parameter can represent a physical property of the system, such as mass, viscosity, or temperature. It can also represent an environmental condition, such as pressure or flow rate. Parameters can also represent model assumptions, such as the geometry of the system or the type of forces acting on it.
What are parameters in mUQSA?
In the context of mUQSA, a parameter is a variable or input to a model that can be varied within a certain range accordingly to some distribution. The set of such parameters constitutes a multidimensional space defining the input uncertainty of the model. In order to perform uncertainty quantification and sensitivity analysis of the model, the parameter space is sampled and the propagation of uncertainty is measured in accordance with one of available methods.
Each parameter in mUQSA is characterized by its name, type, distribution and lower and upper bounds. The name is a string that identifies the parameter, while the type specifies the data type of the parameter (e.g., integer, float, or string). The distribution defines the probability distribution of the parameter values within the parameter space. Finally, the lower and upper bounds are hard limits for the value of certain parameter. mUQSA provides a number of built-in distributions, including uniform, normal, triangular, etc.
mUQSA supports several algorithms for the sampling with the Quasi Monte Carlo, Stochastic Collocation and Polynomial Chaos Expansion methods. These algorithms, based on input distributions and defined limits, generate a set of parameter combinations to the computational model. Next, a model is evaluated as many times as the number of generated combinations. In a final step the output data from the evaluations is collected for the analysis.
When carrying out UQ and SA of a computational model, a useful may be to conceptually distinguish between the regular input parameters and the model’s configuration parameters. Unlike an input parameter which is typically sampled from a relatively wide range of values, the configuration parameter is fixed to specific value or its changeability is highly reduced. This diversification allows us to focus on input variability without losing the possibility to explore how different options for infrequently changed model settings affect the results.