Using covariance weighted euclidean distance to assess the dissimilarity between integral experiments

Integral experiments especially criticality experiments help a lot in designing either new nuclear reactor or criticality assembly. The calculation uncertainty of the integral parameter which is introduced in by the nuclear data uncertainty is larger than the experimental uncertainty for most high-enriched uranium metal experiments, therefore the integral experiment is still very useful. There are lots of integral experiments have been done and documented. It should be considered carefully that which integral experiments should be used in applications. For instance, if the aim of the application is to validate the criticality design of a new reactor, integral experiments which are similar to the new reactor should be used. There are several similarity measures which have been used to assess the similarity between integral experiments, such as E similarity measure, G similarity measure and C similarity measure. But, there is no standard rule to choose which similarity measure should be used to assess the similarity between integral experiments in specific application. Another shortage of these similarity measures is that the thresholds of these similarity measures which should be set to judge whether the integral experiments are similar to each other or not have no clear physical meaning. In this paper, we will analyze the existing similarity measures which have been used to assess the similarity between integral experiments, and test some other similarity or dissimilarity measures which have been used in other research fields. After testing the Tanimato similarity measure and Euclidean distance, we find that the covariance weighted Euclidean distance is well suit to assess the dissimilarity between integral experiments, and the physical meaning of its threshold is clear. We recommend using covariance weighted Euclidean distance to assess the dissimilarity between integral experiments.