The need to be able to model variability at different scales is fundamental for the graphite cores and QuinStats incorporates Quintessa’s approach to handling this variability, which could be applied to any system where variability (at one or more scales) is important. QuinStats also has the capability to undertake uncertainty analyses in a number of different ways.
Before QuinStats was available various approaches were employed to undertake statistical modelling, including simple spreadsheets, the R statistics package and specific dedicated codes. Difficulties had been experienced in fitting non-linear multi-level (hierarchical) models using other general-purpose statistical modelling packages. The flexibility to use different solution methods in different situations enables QuinStats to be used to find solutions for problems that can sometimes be problematic for general-purpose statistical modelling packages.
Fitting a statistical model to the available data is only the first step. Generally, the reason for fitting the model is to allow predictions or forecasts to be made. We use these terms to mean slightly different things. A prediction is made with the intention of comparing it against an observation as part of model validation. Longer-term forecasts are needed for a number of different purposes, particularly to support safety cases for aspects that cannot be directly measured. QuinStats allows variability and uncertainty assessments to be carried through a chain of forecasts as required. The figure below shows an example of forecasts produced using QuinStats. The maximum likelihood estimate (MLE) is shown along with the 95th percentile which quantifies the impact of uncertainty.
An important feature of QuinStats is that the models which it uses are directly specified by the user in the input file. This allows a variety of models to be explored quickly and efficiently. This has also enabled QuinStats to be used to verify calculations made by other dedicated codes.
QuinStats uses pre-existing Quintessa software components together with specific code implementing the required statistical modelling algorithms, enabling the software to be developed rapidly and efficiently.