Intelligent Identification of Boiling Water Reactor State Utilizing Relevance Vector Regression Models

Abstract

Modernization of reactor instrumentation and control systems is mainly characterized by the transition from analog to digital systems, expressed by replacement of hardware equipment with new software-driven devices. Digital systems may share intelligence capabilities where except for measuring and processing information may also make decisions. State identification systems are systems that process the measurements taken over operational variables and output the state of the reactor. This paper frames itself in the area of control systems applied to state identification of boiling water reactors (BWRs). It presents a methodology that utilizes machine learning tools, and more specifically, a set of relevance vector machines (RVMs) in order to process the incoming signals and identify the state of the BWR in real time. The proposed methodology is comprised of two stages: in the first stage, each RVM identifies the state of the BWR, while the second stage collects the RVM outputs and decides about the real state of the reactor adopting majority voting. The proposed methodology is tested on a set of real-world BWR data taken from the experimental FIX-II facility for recognizing various BWR loss-of-coolant accidents (LOCAs) as well as normal states. Results exhibit the efficiency of the methodology in correctly identifying the correct state of the BWR while promoting real time identification by providing fast responses. However, a strong dependence of identification performance on the form of kernel functions is also concluded.