Abstract
The current order-of-magnitude uncertainty associated with scaling factor (SF) and power law (PL) inventory estimates of difficult-to-measure (DTM) radionuclides in radwaste can be significantly reduced by using a correlation based on an artificial neural network (ANN). The linear SF and the non-linear PL correlations use a single γ-emitting (marker) radionuclide to predict concentrations of key, DTM radionuclides in the waste. The newly developed method uses an ANN to produce a non-linear correlation between 2 or more marker radionuclides and the DTM radionuclides. Non-numeric waste characterization data (e.g., waste stream identification) can also be included in the ANN-based correlation to further reduce the uncertainty in the DTM inventory estimates.
