High Voltage Direct Current (HVDC) can quickly vary its power to stabilize an AC power system when the latter is subject to a disturbance. Typically, linear control is used to design that control strategy. However, due to the uncertainties of the faults and the system model, it is necessary to estimate the linear model of the post-fault power system using real-time measurements. This leads to two main problems: Firstly, after the fault, the HVDC will quickly evolve to steady-state operating conditions and, therefore, result in an unobservable control matrix. Secondly, the HVDC power should not exceed its stable range, which is limited by the post-fault system. To address the above issues, the main novelties are as follows: First, this paper proposes a novel system identification method to handle that case when there is no control input excitation. Second, this paper presents an AC/DC hybrid power system control strategy that fully considers the real-time stability constraints of the HVDC. More specifically, this paper proposes to estimate the system matrix and the control matrix separately. Based on the linearized model of the AC/DC hybrid system, the control matrix is calculated using the sensitivities of the active powers of the generators with respect to those of the HVDC. Then, the system matrix is estimated in real-time using measures of the power angle and frequency. Finally, by processing the local voltage and current measurements, the associated Thevenin equivalent parameters on the system side of the HVDC connecting point are estimated in real-time and the upper bound of the DC power is determined. According to the linearized model and the HVDC power constraints, a control strategy is proposed using the model predictive control method. Simulation results reveal the effectiveness of the proposed strategy.