TY - GEN
T1 - A multivariate operational situation awareness method based on weighted graph structure for complex nuclear power systems
AU - Zhang, Shuo
AU - Cheng, Wei
AU - Zhang, Le
AU - Chen, Xuefeng
AU - Chang, Fengtian
AU - Hong, Junying
AU - Ma, Yingfei
AU - Xu, Zhao
AU - Yang, Ruzhen
N1 - Publisher Copyright:
© 2023 SPIE.
PY - 2023
Y1 - 2023
N2 - Operational situation awareness is a key technology to ensure the operation safety of complex nuclear power systems. The existing parallel monitoring method cannot sense latent abnormalities and the trend of abnormal evolution. Simultaneously, the monitoring data of nuclear power systems are characterized by multi-source, high-dimensional, strong coupling, etc. The operational situation of a single variable is affected by other variables to varying degrees, and its correlation cannot be quantitatively evaluated. This paper proposes a multivariate operational situation awareness method for complex nuclear power systems based on a weighted graph structure. First, WGS is used to learn to characterize the dependencies and weights among complex multivariate data and construct an interpretable graph structure. Second, attention-based LSTM is used to realize the measurement point of operational situation awareness. Ultimate, the method is validated on the operational monitoring data of the main and auxiliary systems of a nuclear power unit in one circuit. Compared with ATT-LSTM, WGS-LSTM's computational accuracy and speed are improved by 71.43% and 11.78%, respectively, which can effectively sense the operational situation of complex nuclear power systems.
AB - Operational situation awareness is a key technology to ensure the operation safety of complex nuclear power systems. The existing parallel monitoring method cannot sense latent abnormalities and the trend of abnormal evolution. Simultaneously, the monitoring data of nuclear power systems are characterized by multi-source, high-dimensional, strong coupling, etc. The operational situation of a single variable is affected by other variables to varying degrees, and its correlation cannot be quantitatively evaluated. This paper proposes a multivariate operational situation awareness method for complex nuclear power systems based on a weighted graph structure. First, WGS is used to learn to characterize the dependencies and weights among complex multivariate data and construct an interpretable graph structure. Second, attention-based LSTM is used to realize the measurement point of operational situation awareness. Ultimate, the method is validated on the operational monitoring data of the main and auxiliary systems of a nuclear power unit in one circuit. Compared with ATT-LSTM, WGS-LSTM's computational accuracy and speed are improved by 71.43% and 11.78%, respectively, which can effectively sense the operational situation of complex nuclear power systems.
KW - Graph structure learning
KW - LSTM network
KW - Multivariate time series
KW - Nuclear power safety
KW - Operational situation awareness
UR - https://www.scopus.com/pages/publications/85176497707
U2 - 10.1117/12.2684953
DO - 10.1117/12.2684953
M3 - 会议稿件
AN - SCOPUS:85176497707
T3 - Proceedings of SPIE - The International Society for Optical Engineering
BT - Fourth International Conference on Artificial Intelligence and Electromechanical Automation, AIEA 2023
A2 - Wen, Fushuan
A2 - Zhao, Chuanjun
A2 - Chen, Yanjiao
PB - SPIE
T2 - 4th International Conference on Artificial Intelligence and Electromechanical Automation, AIEA 2023
Y2 - 10 March 2023 through 12 March 2023
ER -