Short-term prediction of time series using radial basis function network incorporating affine transformation and application to control

Abstract

In this paper, we apply a short-term prediction method using an RBF network to conventional feedback control. The conventional short-term prediction method is fixed on the absolute coordinate system and the prediction accuracy is insufficient. In this paper, in order to improve the calculation accuracy without increasing the amount of calculation, the coordinate transformation by affine transformation is introduced. Therefore, the training data on the absolute coordinate system is transformed into the rotating coordinate system using the affine transformation, and short-term prediction using the RBF network is performed on this rotating coordinate system. As a result of numerical calculation, it was shown that the method incorporating the affine transformation improves the prediction accuracy. Comparing the conventional feedback control and the proposed method with the spring mass model of the one-degree-of-freedom system, it was found that the proposed method converges faster.