Neural network-based prediction of decommissioning costs for SMRs

Decommissioning Small Modular Reactors (SMRs) poses distinct economic challenges compared to conventional nuclear power plants. This study uses a statistical technique, Principal Component Analysis (PCA) to reduce collected nine cost-driving data to two principal components, that explain 77.87% of the total variance. They were put into a Radial Basis Function (RBF) neural network as input to develop a predictive model for decommissioning costs, achieving a training error of 0.0016 and demonstrating strong predictive accuracy across 30 sample datasets. Based on result, dismantling the reactor pressure vessel (RPV) is a significant fixed cost, it is the largest contributor of overall expenses. Simulations show that increasing net electrical output (NEO) from 50 MWe to 350 MWe increases total decommissioning costs by only 6%, indicating the independence, and signing the importance of larger scale or co-located installations. This leads to the statement that targeted strategies are needed to optimize dismantling processes and waste management to achieve cost savings by taking the best use of modular design.