Neuroidentification and neurocontrol of dynamic systems using backpropagation technique

In this paper, the technique of forward network modeling of dynamic systems as well as the forward network model based (FNMB) control have been presented. It has been shown by simulation that the FNMB controller outperforms a well tuned conventional PI controller, it is good not only at command following but also at regulation against external disturbances, either deterministic or stochastic; moreover, it has strong robustness to large parameter variations in the controlled system. The FNMB controller has also been shown to be able to perform almost perfect decoupling control over a highly coupled nonlinear multi-input multi-output (MIMO) dynamic system when the command magnitude is low enough. The unique bursting phenomenon appeared in the MIMO case as well as the problem concerned with training data collection needed for modeling dynamic systems with networks are discussed at the end of the paper