Parameter studies involve systematically varying input parameters of a model to analyze their effects on the system’s behavior or results. These studies are crucial for understanding the influence of the model to different parameters, optimizing performance, and identifying critical thresholds or trends. Exploring a range of parameter values gives insights into the stability, accuracy, and robustness of numerical methods and computational models, enabling better predictions and design decisions in engineering and scientific applications.
Features in QUEENS
- Grid Sampling: This involves creating a grid of parameter values and evaluating the model at each grid point. It is straightforward but can become computationally expensive for high-dimensional parameter spaces.
- Design of Experiment (DoE): DoE approaches systematically explore the parameter space by selecting a representative set of sample points to maximize information gain while minimizing computational effort. Methods like Latin Hypercube Sampling (LHS), factorial designs, and space-filling designs enable efficient analysis of parameter effects, interactions, and optimization within complex models.
- User-defined List of Points: The user-defined list of points approach allows users to explicitly specify the sample points in the parameter space based on prior knowledge or targeted exploration goals. This method offers complete flexibility to focus on specific regions of interest, test hypotheses, or evaluate irregular or non-uniform patterns, making it ideal for customized studies and scenarios with computational or experimental constraints.