Adaptive Nonlinear Dynamic Inversion Control using RBF Neural Network

摘要

Most currents Adaptive Nonlinear Dynamic Inversion (ANDI) methodologies employing Neural Networks (NN) cancel the error due to approximate inversion. But since the characteristics of the actuator saturation, the NN will attempt to adapt to these characteristics even when it might not be desirable to do so. Pseudo Control Hedging (PCH) is one of the more recent methods introduced to address this issue. PCH requires the system be controlled to track a reference model.What's more, this paper analysed the inverse error. Learning of Radial-Basis-Function (RBF) NN was accomplished by a simple weight update rule derived from Lyapunov theory, thus assuring the stability of the closed-loop system. All these are tested by simulation. The simulating results in a design that is robust with respect to aerodynamic modeling inaccuracies and to external disturbances.