Reduced Order Model Implementation in the Risk-Informed Safety Margin Characterization Toolkit

Enhancements to the RAVEN code in FY16

“…Many methods provide point wise, local sensitivity, which is more amenable to drill-down tasks such as optimization rather than the summarized information we seek to provide. Such methods can either not provide enough information when the visualization is focused around a particular location [5,8,54] or create unnecessary clutter when trying to extract trends in the data [13,14,30].…”

“…HyperMoVal [54] allows users to visually validate and explore a model built using support vector regression (SVR) [58] against real data in order to identify and understand regions of poor fitting. Berger et al [5] focus instead on local optimization by providing local sensitivity information on and around a focal point in a focus+context visualization. Vismon [8] uses sensitivity information to perform trade-off analysis.…”

“…As already mentioned in previous sections and detailed in previous reports [5,3], an important pathway in RAVEN analysis workflow is represented by the usage of Surrogate Models (a.k.a. Reduced Order Model -ROM).…”

“…For a fixed scale, the main idea behind the simplification is to merge its corresponding partitions based on a measure of their significance (i.e., persistence); see Figure 3 below for an example applied to the MSC of a 2D height function. For point cloud data, the MSC can be approximated [5,14], enabling MSR to be applied in high dimensions.…”

Light Water Reactor Sustainability Program Status of Adaptive Surrogates within the RAVEN framework

“…The HybridModel is designed to combine multiple surrogate models and any other Model (i.e. high-fidelity model) leveraging the EnsembleModel infrastructure developed in FY15 [ 9 ] and improvement last year [10], deciding which of the Model needs to be evaluated based on the model validation score.…”