Training Neural Networks for Robust Control of Nonlinear Mimo Systems

A training strategy for computational neural networks is introduced that paves the way for incorporation of neural networks in robust control design for nonlinear multiple input, multiple output systems. The proposed training strategy enables utilization of statistical properties of the least-squares estimate. A control strategy that has a structural similarity to an adaptive control structure is adopted and it is outlined how neural networks which are trained with the proposed training strategy can be used to incorporate robustness in this control strategy.