Probabilistic Risk Assessment

In response to a congressional request, GAO reported on: (1) the state of the art of the Nuclear Regulatory Commission's (NRC) probabilistic risk assessment (PRA); (2) whether the NRC use of PRA appears reasonable considering its staff's experience and training; and (3) whether PRA adequately considers the potential problems and disadvantages of the analysis method.

GAO has found that many improvements have been made in PRA methodology since it was first used in 1975; however, uncertainties remain because PRA identifies and assigns probabilities to nuclear accident events that rarely occur. The uncertainties also reflect the incomplete knowledge about plant systems, human behavior, accident processes, the off-site consequences of accidents, and how external events can cause accidents. Therefore, due to insufficient and unreliable data, analysts may make poor assumptions, and computer models may not be realistic. In 1983, NRC began a 3-year research program to reduce some of these uncertainties and to develop a computer model, collect experimental and actuarial data, improve its models, improve its understanding of accident processes, and develop models and data on external events. NRC uses PRA to analyze: (1) nuclear power plants and plant systems; (2) related regulations and safety issues; and (3) the estimated costs and benefits of alternative regulatory actions. Although the use of PRA is costly and time-consuming, increased staff and contractor training and experience have made its use timely and reasonable. However, NRC should not use the numerical risk estimates as the sole or primary basis for regulatory decisions and should use PRA to supplement its more traditional analytical and engineering methods.