Autoencoder-Based Model of Nuclear Power Plant Core Temperature for Blockage Event Detection

This paper addresses the critical necessity of continuously monitoring systems of the temperature of fuel rod assemblies within a nuclear power plant and presents a quick and accurate detection of total and instantaneous blockages. Our research works propose an advanced and original autoencoderbased methodology for adaptive modelling of normal operational temperatures. This model allows incorporating detailed expertise on the specific malfunctions targeted, in order to detect and identify small heat increases indicative of blockages in the flow of a nuclear core coolant. The proposed online detection method ensures a reliable outcome in terms of false alarm probability as well as detection delay which are both of the utmost importance. Experimental results, utilizing real actual temperature measurements from the Superphénix power station, demonstrate the model’s accuracy and the relevance of the proposed detection method.