Abstract: Because the conventional torque control method can not be used to build up an accurate mathematical model,in order to improve the efficiency and reliability of online parameter identification for torpedo asynchronous motor,a method for identifying the torque of a torpedo AC motor based on back propagation (BP) neural network is presented to meet the requirement of system performance. An equation of AC motor torque is derived on hypothesis,and the torque is identified by the BP neural network trained online or offline, without the need for a mathematical model of target.The training results are used to design the identifying system.The simulation results show better accuracy of the torque identification with the proposed method.