An Accident Diagnosis Algorithm for HTR-PM Based on Deep Learning Methods

Zeguang Li,Jun Sun,Jiejuan Tong,Zhe Sui,Langming Gang

DOI: https://doi.org/10.1016/j.pnucene.2019.03.038

IF: 2.461

2019-01-01

Progress in Nuclear Energy

Abstract:The Chinese High Temperature Reactor Pebble-bed Module (HTR-PM) is the world first commercial nuclear power plant (NPP) with the characteristics of fourth generation. Accident diagnosis tasks for HTR-PM are directly associated with safe and efficient operation. Although different kinds of accident diagnosis methods have been studies on conventional NPPs, the research of accident diagnosis of HTR-PM is relatively lacking according to the different characteristics and new applications of HTR-PM. In this article, a new algorithm for HTR-PM accident diagnosis based on deep learning methods is proposed. By using the preprocessing, classification network and postprocessing techniques, the proposed algorithm could avoid over-reliance on the previous experiences and make fully use of the signals, also it can use only few number of training signal sequences to get high accuracy results, which is a significant advantage compared with traditional algorithms using deep learning methods. This algorithm is tested using the signals produced by the engineering simulator of HTR-PM for normal state and several accidents, including loss of feed water, large break depressurized loss of forced cooling, small break depressurized loss of forced cooling and inadvertent withdrawal of a single control rod. The results show the feasibility and effectiveness of the algorithm for HTR-PM accident diagnosis, and also the potentiality to use in other NPPs accident diagnosis tasks.