Efficient self-learning disturbance-resistant control for high-speed flight vehicle based on dual heuristic dynamic programming

• Published in: Engineering applications of artificial intelligence Vol. 150; p. 110521
• Main Authors: Huang, Xu, Liu, Jiarun, Peng, Yue, Zhang, Yuan, Wang, Zhaolei, Bao, Weimin
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 15.06.2025
• Subjects: Adaptive dynamic programming; Attitude control; Critic-only structure; Disturbance-resistant control; Neural network; Critic-only structure; Attitude control; Neural network; Adaptive dynamic programming; Disturbance-resistant control
• ISSN: 0952-1976
• DOI: 10.1016/j.engappai.2025.110521

Recent advancements in high-speed flight vehicles (HSFVs) have sparked significant interest due to their strategic importance and emerging civilian applications. These vehicles exhibit strong nonlinearities and multi-axis interactions and are usually influenced by uncertainties such as modeling errors, parameter perturbations, and external disturbances. Neglecting these challenges in attitude controller design can lead to trajectory tracking deviations and potential mission failure due to instability. Motivated by this issue, an efficient disturbance-resistant control method with online self-learning capability is proposed. Firstly, a feedback linearization baseline controller combined with finite-time extended state observers (FESOs) is designed to ensure stability. Next, a dual heuristic dynamic programming (DHP) controller with critic-only structure is developed for online performance optimization. Update laws of the critic neural network (NN) are derived based on policy iteration, and zero-sum game (ZSG) theory is incorporated to enhance the system’s adaptive capacity to uncertainties. Lyapunov theory is subsequently employed to validate the convergence of network weights and the system stability. The proposed method, compared to common adaptive dynamic programming (ADP) approaches for attitude control, demonstrates superior learning efficiency and guarantees the convergence of online learning without the necessity for pre-training. Simulation results indicate that the method equips the HSFV with robust dynamic performance throughout a broad flight envelope, with attitude tracking errors constrained to less than 0.5°. Future research will focus on developing fault-tolerant and prescribed performance control frameworks with online learning ability, representing an advancement in the current technique. [Display omitted] •A self-learning control framework is proposed for three-axis attitude control of HSFVs.•Novel update laws for policy iteration DHP expedite online learning convergence.•The framework, with low time and space complexity, facilitates implementation on flight control computers.