In view of the fact that the armoured vehicle carrying obstacle-breaking rocket weapon may vibrate on a rough road, while a large vibration will also be generated when launching obstacle-breaking shells, which may easily cause the deviation of the direction angle of the next obstacle-breaking and affect the obstacle-breaking accuracy. With consideration of other disturbances and uncertainties, the obstacle-breaking weapon system can be taken as a nonlinear time-varying system. Based on the robustness and adaptability of fuzzy control and the adaptive and self-learning ability of neural network, a PID control algorithm based on fuzzy RBF neural network is proposed. At the same time, the structure parameter values of fuzzy neural network are initialized by K-means hierarchical clustering, and the fuzzy neural network is trained by using the LM algorithm. The simulation results show that the method can effectively improve the anti-interference ability, the accuracy of the obstacle-breaking and the speed of the gun adjustment.