Adaptive dynamic programming for ramjet intelligent tracking control via neural network-enhanced equilibrium manifold expansion estimator

Precise control of the operational state of turbine-based combined cycle (TBCC) high-speed channel is a crucial factor in ensuring the reliable implementation of the mode transition (MT). To achieve this goal, offline training and online updating of the control-oriented model (COM), neural network (NN), and reinforcement learning are considered in this study. First, the Mach number (Ma) is introduced into the NN-enhanced equilibrium manifold expansion (NNEME) model to characterize the wide-speed-range operational characteristics of the ramjet engine. Furthermore, the extended state observer (ESO) is employed to further estimate unmodelled dynamics, and the adaptive dynamic programming (ADP) error controller is established to address the control problem of the ramjet engine during MT process. Throughout the entire MT process, the error confidence intervals for π₄₀ and π₃₂ output by the NNEME model are reduced by 92.97 % and 93.66 %, respectively. Moreover, hardware-in-the-loop (HIL) results indicate that the ESO + NNEME ADP control system is more stable and reduces state overshoots by up to 56.9178 % and 82.8062 %, respectively. The proposed advanced control system composed of ESO + NNEME steady-state controller and ADP error controller exhibits a better control performance and robustness within the entire MT process.