Prediction Method of a System with Continuous Change Parameters Considering SNR Fluctuation

Abstract

This paper presents a new prediction method for time-varying parameter systems. This method is based on on-line estimation approximating continuous changes of system parameters by neural networks. However, since sufficient prediction accuracy can not be obtained with the previous method, an error function that considered a signal-to-noise ratio fluctuation and time-weight was used. The predictive values of time-varying parameters is sufficiently obtained by the learning ability and generalization ability of neural networks. According to the simulations, it has been confirmed that the proposed prediction method can obtain accurate prediction values compared with the conventional parameter prediction.