Experimental evaluation of neural nonlinear modeling techniques

摘要

In this paper three nonlinear neural modeling techniques, multi-layer perceptron networks (MLP), B-spline networks, and radial basis function networks (RBFN), are compared in an experimental application for a laboratory scale pH neutralization process at UCSB. The neural modeling techniques are evaluated on the basis of model accuracy, practicality, and ease of implementation. It will be shown that accurate models can be otained using any of the networks, and that RBFN and B-Spline networks have the benefit of local basis function support. Furthermore, the effects of the "Curse of Dimensionality" on the three networks typies will be discussed in terms of computational difficulty and practicality.