Abstract: A covert attack method based on a symbiotic organism search(SOS) algorithm to optimize long short-term memory (LSTM) neural network was proposed to solve the problem of obtaining a high-precision estimation model of the attacked target for covert attacks. The output and input signals of the feedback controller of the attack target were taken as the data set of the LSTM. The estimation model of the attacked area was obtained through training, and was used to design the covert attacker to impose attack signals on the attacked object.In addition, the SOS algorithm was applied to optimize the parameters of the LSTM to improve the performance of the covert attacker.The simulation results of covert attack on the primary circuit control system of nuclear power plant show that the attack method has high concealment performance while realizing preset attack behavior on the output signal of the target control system.