Research on fine management of abnormal response mode of nuclear power plant in group reactor mode

Abstract This study presents a data-driven framework for controlling abnormal responses in multi-reactor nuclear power plants under cooperative operating conditions. By integrating multiple-source operational parameters – including thermal power, steam pressure, coolant flow rates, and structural pressure dynamics – the system effectively models and identifies transient safety risks, such as pressurised vessel damage, molten relocation, and steam leakage. A novel Asymmetric Multi-Classifier Learning (AMCL) approach is proposed to enhance classification accuracy in the presence of class imbalance and dynamic uncertainty. Experimental validation using internal iterative loss convergence curves, confusion matrices, and real plant incident recordings demonstrates the superiority of the proposed method over traditional STL, SMCL, and balanced-AMCL schemes. Furthermore, multi-scenario response graphs confirm the system’s adaptability across typical load disturbances and emergency conditions.