Data-driven preview Q-learning for H_∞ tracking control of unknown linear discrete-time systems

Conventional tracking controllers relying only on the current tracking error may suffer from tracking lag when the reference trajectory changes rapidly. To address this problem, a data-driven preview Q-learning algorithm is presented for H_∞ tracking control of unknown linear discrete-time systems. The algorithm integrates preview information, enabling the controller to utilize known future reference variations and adjust the control action in advance. By incorporating the tracking error and preview information, an augmented error system is constructed. A corresponding state reconstruction relation is then established, allowing the Q-function to be reformulated using measurable input-output and disturbance data together with preview information. Based on this formulation, the data-driven preview Q-learning algorithm is developed to solve the tracking problem without requiring prior knowledge of the system model, full state information, or an initially admissible policy. Simulation results demonstrate that the proposed method achieves higher tracking accuracy, improved transient tracking performance, and satisfactory robustness under noisy external disturbances.