Annealing Robust Radial Basis Function Networks for Modeling with Outliers

Abstract

In this paper, the annealing robust radial basis function networks (ARRBFNs) is proposed to improve the problems of robust RBFNs for function approximation with outliers. Firstly, the support vector regression (SVR) approach is used to obtain the initial structure of ARRBFNs. Because of the SVR approach is equivalent to solving a linear constrained quadratic programming problem under the fixed structure of SVR, the number of hidden nodes and adjustable parameters (e.g. initial structure) are easy obtained in the ARRBFNs. Secondly, we use the results of SVR as initialization of ARRBFNs. Then, the annealing robust backpropagation (ARBP) learning algorithm used as the learning algorithm of ARRBFNs and applied to adjust the parameters of ARRBFNs. The ARBP learning algorithm has been proposed to overcome the problems of initialization and cut-off points in the robust learning algorithm. Based on the initialization of ARRBFNs by SVR approach, the ARRBFNs have a fast convergence speed and robust against outliers. Simulation results are provided to show the validity and applicability of the proposed ARRBFNs.