A nonparametric neural networks for adaptive real-time control

This paper presents a nonparametric artificial neural network (ANN) model for adaptive control of nonlinear systems. The proposed ANN, a functional polynomial network (FPN), mixes the concept of orthogonal basis functions with the idea of polynomial networks. A combination of orthogonal functions can be used to produce a desired mapping. However, there is no way besides trial and error to choose which orthogonal functions should be selected. Polynomial nets can be used for function approximation, but, it is not easy to set the order of the activation function. The combination of the two concepts produces a very powerful ANN model due to the automatic input selection capability of the training procedures employed by polynomial networks. The proposed FPN has been tested for speed control of a DC motor. The results have been compared with the ones provided by an indirect adaptive control scheme based on multilayer perceptrons trained by backpropagation