Emulation of probabilistic-safety-assessment code in a nuclear power plant risk monitor by artificial neural networks

This work represents a feasibility study of utilizing artificial neural networks for quick evaluation of Probabilistic Safety Assessment (PSA) models in the risk monitors of nuclear power plants.

The study provides a brief overview of methodology and applications of the PSA, and a detailed description of the tasks and structure of the risk monitors and their individual parts. It identifies a possible role of the artificial neural networks in the risk monitors and gives an overview of the paradigms suitable for this application. The approach used and the results obtained when using these paradigms for evaluation of a small test PSA model and their expected performance for the full-scope PSA models are discussed in detail.

The results obtained demonstrate that though the artificial neural networks would be probably capable of performing the desired evaluation of the PSA models, their training would require pre-computation of such a vast volume of data that it makes their proposed utilization impractical. The study concludes that the investigated application of the artificial neural networks in the risk monitor is not feasible in the present state of knowledge.