Hierarchical optimization-based online learning control for nonlinear systems with unknown disturbances

A hierarchical optimization-based online learning control scheme is presented to exploit the learning control performance of neural networks (NNs) for nonlinear systems with unknown disturbances. First, leveraging the equivalent-input-disturbance (EID) concept and using only the system output, an NN-assisted control structure is devised. Then, a two-stage learning control framework is developed. Specifically, the first rough stage is for fast dynamic disturbance-rejection control with exponential convergence performance and then the second refined stage is to further refine the performance by online optimization. Further, the closed-loop performance with integrated disturbance-rejection performance is analyzed and a performance-oriented design algorithm is presented. Finally, superiorities of the developed scheme are shown by comparisons of rotary inverted pendulum experiments with other methods.