The estimate accuracy of wind disturbance， airflow disturbance and unknown part of the model directly affects the stability and control performance of the flight vehicle. The Extended State Observer （ESO） can estimate wind disturbance， airflow disturbance and unknown part of the model， but there are problems of low type and low estimate accuracy. Compensation Function Observer （CFO） adopts the idea of pure integral， compensation and transfer function type， and changes the structure of ESO， making CFO two types higher than ESO， with high precision and strong convergence. However， CFO uses a linear filter to compensate the unknown function or disturbance of the system， and has insufficient compensation ability for fast-changing or high-order nonlinear functions. Because the unknown model function is often a nonlinear function， in this paper， a Radial Basis Function（RBF） neural network is used to replace the linear filter， and a compensation function observer with RBF neural network is proposed， which further improves the estimation accuracy. Using the nonlinear unknown model and disturbance， and differential information obtained by the compensation function observer with RBF neural network， an active model compensation control algorithm is designed and successfully applied to the control of the quadrotor flight vehicle attitude system. The stability of the closed-loop system is proved by applying Lyapunov stability theory. Through simulations， the proposed model compensation control based on the compensation function observer is compared with the PID control and active disturbance rejection control algorithm. On the Pixhawk-based control test platform experiment， the quadrotor flight vehicle is tested using these three control strategies for different tracking performance for reference poses. The results show that the proposed control method substantially outperforms other controllers in transient performance and steady-state tracking accuracy.