Nonlinear prediction using radial basis function network incorporating coordinate transformation

抄録

A nonlinear modeling and prediction method using radial basis function (RBF) network is developed in this paper, where the coordinate system of training data is transformed using the affine transformation. It is difficult to accurately predict the time series by conventional approach using the RBF network, in which the coordinate system is always fixed and this makes the prediction poor. For highly accurate prediction using the RBF network, the coordinate system is transformed using the affine transformation. The coordinate system of training data for the RBF network is transformed, and consequently it is expected that the highly accurate prediction can be made in comparison with the conventional approach. The proposed approach is numerically easy to implement with the computationally inexpensive procedure. The validation and characteristics of the proposed approach are examined through benchmarks.