This paper presents a feedforward-feedback control system based on the double neural network. The operation data of the single gimbal control moment gyro（SGCMG） system under different conditions is collected and processed by Kalman filter. A virtual generalized controlled object model covering the information of the control channel and the disturbance channel is constructed based on neural network， and a neural network feedforward compensator is designed. The composite control system of feedforward compensator and original feedback controller is realized， which reduces the constraint that the traditional feedforward-feedback composite control needs to obtain the disturbance channel information accurately， and can overcome the influence of the complex disturbance existing in the SGCMG system on the control performance. The simulation results show that the scheme can improve the steady-state accuracy of the system， reduce the steady-state error and improve the control dynamic performance.