<p>Hypersonic vehicles face significant challenges in control system design due to strong uncertainties in complex flight environments. To address this, this paper proposes an intelligent control method based on a modified deep deterministic policy gradient algorithm. First, to overcome the limitations of traditional reward functions in deep deterministic policy gradient, a segmented reward function is designed that integrates tracking error, error rate of change, and control input. This design more effectively guides agent learning, ensures tracking accuracy, and significantly improves control smoothness and convergence speed. Second, the improved algorithm is used to adaptively tune the parameters of the extended state observer in the altitude subsystem online, overcoming the tendency of traditional optimization methods to fall into local optima and enhancing the system’s adaptability to uncertainties. Furthermore, the proposed modified deep deterministic policy gradient-based sliding mode active disturbance rejection control system demonstrates excellent robustness under various complex scenarios, including parameter perturbation, external disturbances, and actuator faults. Compared with traditional deep deterministic policy gradient optimization approaches, the proposed method shows clear advantages in disturbance rejection and fault tolerance: Under parameter perturbations and external disturbances, the comprehensive performance index of altitude tracking improves by approximately 50% on average; under actuator faults, all performance metrics are also comprehensively improved.</p>