Nonlinear Dynamical Decoupling Controller using Multiple Neural Network Models

A multiple models neural network decoupling controller is designed to control the multivariable nonlinear discrete time system. First, at each equilibrium point, one neural network is trained offline to identify the linear term of the nonlinear system while the other neural network is trained online to identify the nonlinear term. Then the multiple models is composed of all models which is got from each equilibrium point. According to the switching index, the best model is selected as the system model. So, for the selected model, the nonlinear term is viewed as the measurable disturbance and eliminated using feedforward strategy. By the choice of the polynomial matrices, the poles of the linear term can be placed arbitrarily and the dynamical decoupling controller is designed accordingly. The simulation example shows effectiveness of the proposed system