Use of neural networks for direct self-tuning control of stochastic no
nlinear plants has been proposed. The control is based upon inverse mo
delling of a pseudo-plant. The input to the pseudo-plant is same as th
e plant input while its output consists of a linear combination of the
plant input and output. The controller is directly identified as a me
an square optimal inverse estimator of the pseudo-plant. This approach
allows the control of inverse unstable plants. Local convergence prop
erties as well as results of simulation studies are presented.