Abstract: In order to improve the accuracy of gyro constantcurrent source, a  compensation method of gyro constantcurrent source is proposed based on back propagation (BP) neural network. Firstly, with BP neural network, a nonlinear steadystate model that represents the relationship between the control input and the output of gyro constantcurrent source is trained. The model is used to estimate the output deviation of constantcurrent source. Secondly, a compensation criterion is confirmed, which is used to determine whether or not to modify the control command. Once the constantcurrent output deviation exceeds the margin, a certain proportion deviation is added up to the control command. Finally, the HIL(hardwareintheloop) simulation is carried out. The results show that the proposed method improves the output accuracy of the gyro constantcurrent greatly compared with the conventional piecewise linear correction method.

Key words: constantcurrent source, BP neural network, nonlinear model, correction method