Towards a stability and approximation theory for neuro-controllers

This paper discusses issues related to the control of systems using neural networks. The importance of convergence speed in adaptive systems is observed leading to the use of approximation functions such as the radial basis function (RBF) rather than back propagation. Initial trials have been conducted using RBFs for system identification and the results are reported. Attention has been given to the choice of number and location of RBF centres, both of factors having significant influence on the network performance. The aim of the research is to establish stability for certain classes of neuro-controller and initial thoughts are recorded here.<>