Proceedings of the Eighth International Conference on Probabilistic Safety Assessment & Management (PSAM)
22 Using Uncertainty Information in Level 3 PRA Calculations (PSAM-0233)
Download citation file:
- Ris (Zotero)
- Reference Manager
The U.S. Nuclear Regulatory Commission (NRC) uses off-site consequence estimates for Level 3 Probabilistic Risk Assessments (PRAs), for cost-benefit analyses, and for planning purposes. The code that NRC uses for these calculations is MACCS2, MELCOR Accident Consequence Code System, Version 2. MACCS2 accepts source term, weather, and other inputs and estimates doses to populations, early health effects, latent health effects, and economic consequences, including the pathways of cloudshine, groundshine, inhalation, skin deposition, and the food chain.
Routine quantification of uncertainty is of increasing importance. To further this objective, the NRC has added capability to input ranges of values and degrees of belief for most floating point variables in the MACCS2 input stream
A pre-processor executes a Latin Hypercube Sampling (LHS) code to prepare a user-specified number of MACCS2 input decks to reflect variations of the uncertain parameters specified. A random sample is drawn from each LHS interval for each variable and the resulting specific values of the variables are randomly combined to produce the MACCS2 input decks. This yields the required number of independent and equally likely MACCS2 cases.
Source term uncertainty presents special challenges. This group of parameters can be made uncertain and treated in the manner described above. However, this process can be complex. To alleviate this problem, an interface between the NRC's Level 2 analysis code, MELCOR, and the Level 3 input has been developed. If a set of MELCOR analyses were conducted using either Monte Carlo or LHS methods to describe the uncertainty, source terms representing independent, equally-likely outcomes for the accident scenario would be available. MACCS2 uses this set of source terms in a cyclic manner. Because the source term inputs and the group of other parameters are in no way ordered, the cyclic use of source term information is unbiased.
The NRC and the Commission of European Communities (CEC) undertook a review of the generic Level 3 PRA parameters that were considered by a group of experts to be the most important. All of the uncertain processes and phenomena evaluated by the NRC/CEC team have been cast into MACCS2 input ranges and degrees of belief. The process has not proved to be straightforward and preparing the input has required different handling techniques. Each of the parameter ranges specified by the experts was resampled, the resampled values were merged and ordered, and the combined distribution used to represent the overall uncertainty in the parameter.
MACCS2 requires many site-specific parameters in addition to the generic parameters discussed above. Many of these may be important in the estimation of the overall uncertainty and they deserve attention in any full-scope uncertainty analysis. However, the treatment of these parameters is beyond the scope of this discussion.
A special consideration of uncertainty in Level 3 PRA analyses is the issue of sampling the meteorological conditions. The importance of sampling the weather was recognized from the days of the Reactor Safety Study. Therefore, MACCS2 provides the user with options for explicit sampling of a site's meteorology. The option most often chosen is “bin” sampling where the binning focuses on hours of precipitation. Random sampling is used to choose start times from the hours within each bin. This departure from strictly random sampling is desirable because precipitation is not frequent, varying between about 1% and 10% of the hours in an average year in the U.S. The binning and sampling process results in a weight that must be applied to the outcomes calculated for each hour that is chosen in the sample.
MACCS2's atmospheric transport and dispersion model is a straight-line Gaussian plume. In fact material in the atmosphere does not move in such a simple way and assuming that it does is a source of uncertainty. Today this source of uncertainty is recognized, but not explicitly evaluated because normally MACCS2 is used only to predict average values. This was recently studied for one situation and shown to be within about a factor of two.
To combine these differing treatments in a manner that remains unbiased, the exact same weather trials are used for each MACCS2 run, each MACCS2 run is assigned a source term in a cyclic fashion, and a random set of all other parameters from the LHS run is assigned.
The needs of Level 3 analysts for uncertainty information should be kept in mind by Level 1 and Level 2 analysts.