TY - GEN
T1 - A machine learning method integrating neural networks and Gaussian processes for LOCA identification in BWR
AU - Alamaniotis, Miltiadis
AU - Ray, Asok
PY - 2019/1/1
Y1 - 2019/1/1
N2 - Monitoring of Boiling Water Rectors (BWRs) is a complex process that requires the use of a numerous sensors and systems. Acquisition of data and the subsequent processing of it accommodate inference making with regard to the state of the reactor system. System identification promotes decision making with regard to operation action taking. In this paper, we present a new method for serially integrating two machine learning tools and more specifically a neural network and a set of algorithms for learning Gaussian processes. Both sets of tools exhibit learning capabilities, and their integration in the current work offers a two-stage learning schema applied to identification of transient states in BWR. In particular, the proposed methodology utilizes the synergism of a set of Gaussian processes with a feedforward neural network for recognizing the type of loss of coolant accident (LOCA) that occurs in the reactor. The methodology is tested on a set of real-world datasets taken from the FIX-II facility. Results demonstrate efficacy of the method to accurately identify the occurring LOCA among three possible states.
AB - Monitoring of Boiling Water Rectors (BWRs) is a complex process that requires the use of a numerous sensors and systems. Acquisition of data and the subsequent processing of it accommodate inference making with regard to the state of the reactor system. System identification promotes decision making with regard to operation action taking. In this paper, we present a new method for serially integrating two machine learning tools and more specifically a neural network and a set of algorithms for learning Gaussian processes. Both sets of tools exhibit learning capabilities, and their integration in the current work offers a two-stage learning schema applied to identification of transient states in BWR. In particular, the proposed methodology utilizes the synergism of a set of Gaussian processes with a feedforward neural network for recognizing the type of loss of coolant accident (LOCA) that occurs in the reactor. The methodology is tested on a set of real-world datasets taken from the FIX-II facility. Results demonstrate efficacy of the method to accurately identify the occurring LOCA among three possible states.
UR - http://www.scopus.com/inward/record.url?scp=85071025312&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85071025312&partnerID=8YFLogxK
M3 - Conference contribution
T3 - 11th Nuclear Plant Instrumentation, Control, and Human-Machine Interface Technologies, NPIC and HMIT 2019
SP - 431
EP - 439
BT - 11th Nuclear Plant Instrumentation, Control, and Human-Machine Interface Technologies, NPIC and HMIT 2019
PB - American Nuclear Society
T2 - 11th Nuclear Plant Instrumentation, Control, and Human-Machine Interface Technologies, NPIC and HMIT 2019
Y2 - 9 February 2019 through 14 February 2019
ER -