Continuous mapping of nuclear reactor core power using artificial neural network even in the presence of inactive detectors

Abstract

Monitoring the radial power distribution during the operation of a pressurized light water reactor (PWR) is crucial for ensuring safe operating conditions and achieving high levels of fuel burnup. This paper introduces a methodology utilizing Artificial Neural Networks (ANN) for reconstructing the radial power distribution in the core of a Pressurized Water Reactor (PWR) with a hot full power of 1876 MWth, such as the Angra 1 reactor. This approach uses measurements from Self-Powered Neutron Detectors (SPND), simulated through the SERPENT code. The use of ANN demonstrated good accuracy in predicting the radial power distribution with an average relative error of 1.27%, considering 36 active detectors, with maximum relative error of 6.99%. Moreover, the proposed process demonstrated robust performance, even when measurements from one, two, or three SPND detectors were unavailable, resulting in errors of 1.24%, 1.13 %, and 1.09%, respectively. Therefore, the methodology ensures a reliable reconstruction of the radial power distribution, even when SPND detector measurements are unavailable, enabling the optimization of detector use and contributing to the increase of operational safety margins.