摘要：AbstractThis paper proposes a composite control approach for a class of semi-coupled two-time-scale systems, whose model parameters are completely unknown. In the framework of adaptive dynamic programming (ADP), a novel model-free off-policy learning algorithm without ill-conditioned numerical issues is derived on the basis of input and state data. The convergence of the developed ADP algorithm is guaranteed under reasonable assumptions. Using singular perturbation theory, we prove that the closed-loop system with the obtained controller in each iteration is asymptotically stable. Simulation results on an industrial process demonstrate the feasibility and effectiveness of our proposed approach.
关键词：KeywordsSingular perturbation theorymodel-free controladaptive dynamic programmingtwo-time-scale systemsoptimal control
