 
Summary: Prwcedingr ot the 1991IEEE ln~ema~ionalSymposium
13  IS Augur, 1991. Arlington. Vqln13. U S A
"I, lntelllgem Control
A GAUSSIAN NEURAL NETWORK IMPLEMENTATION FOR CONTROL SCHEDULING
Michael A. S w r i and PanosJ. Antsaklis
Depamnentof Electrical Engineering
University of Nom Dame
Notre Dame, Indiana 46556
ABSTRAD
Using neurons with gaussian nonlinearities, a neural network is
designed to implement a control law scheduler. For the implementation
discussed here, the neural network is supplied informationabout existing
operating conditions and then responds by supplying control law
parameter values to the controller. The neural network has two layers of
weights, and the values of the weights and biases are based on given
operating points for the scheduler. By designing the neural networks
generalization behavior, specifications for the interpolation between the
given operating points are satisfied. The neural network implementation
performs best when the operating points are equidistant and has some
drawbackswhen used to implement multiparameter schedulers.
