Covert attacks for primary loops of PHWR nuclear power unit based on SOS-LSTM

Abstract

The ongoing advancement of industry and technology has led industrial control systems to evolve towards integration. System structures become increasingly complex, which renders them increasingly vulnerable to external attacks. Due to their clandestine and destructive character, covert attacks pose a significant threat to the secure operation of nuclear power unit control systems. In order to optimize the performance of control systems for nuclear power units, it is important to study the damage process caused by covert attacks on these systems. Facing the problem of obtaining high-precision estimation models of attack targets for covert attacks, this paper proposes a model estimation method based on long and short-term memory (LSTM) neural network and symbiotic organisms search (SOS) algorithm, which takes the feedback controller output and input signals of the attacking target as the dataset of the LSTM neural network, and optimizes the network parameters of the LSTM neural network using SOS algorithm to improve the accuracy of the model, and designs the covert attacker by obtaining the estimation model of the attacked area through training. The root mean square error of the estimation model for the primary loop of the nuclear power unit has been verified by comparative experiments to be reduced by at least 93.59%, 96.52%, and 91.11%, respectively, compared with the other methods. Loop experiment results concerning the covert attack for the primary loop of nuclear power unit illustrate that this attack method successfully meets the predefined objectives while maintaining high levels of stealthiness.