Dynamic Probabilistic Risk Assessment: A Review of Techniques to Reduce Computational Costs

Dynamic probabilistic risk assessment (PRA) is an advanced method to assess the risk posed by the operation of large-scale engineering systems, such as nuclear power plants. Dynamic PRA methods typically estimate system risk using an automated scenario generation process coupled to a system dynamics computer code. Many scenarios are generated, and each scenario’s corresponding system dynamics are simulated, resulting in a (typically) large computational cost for each risk assessment.Partly because of their computational cost, dynamic PRA methods have not been widely adopted. To address this challenge, many recent methods for dynamic PRA have been developed that incorporate various cost reduction techniques with the goal of reducing their computational cost. This paper provides a review of such cost reduction techniques for dynamic PRA, including algorithmic cost reduction techniques and surrogate modeling techniques, and presents perspectives on these existing techniques.The former reduces the computational cost of dynamic PRA by introducing some form of intelligence into the scenario generation process that guides it toward scenarios deemed to be of interest from a risk analysis perspective. The latter has also been applied to dynamic PRA to reduce the computational cost associated with the simulation of each scenario. Each of the algorithmic cost reduction techniques is classified based on its primary operating principle(s) and is described in detail. Surrogate modeling techniques are also classified based on the type of surrogate model developed. This paper then describes gaps in the literature that could be filled to further advance dynamic PRA, including further reduction of its computational cost.